activity recognition alzheimer bioinformatics cheminformatics clustering crossvalidation data mining deep learning density estimation dimensionality reduction external validation graph mining HPC inductive logic programming kernels localization machine learning medicine mobile systems and mobility possible worlds QSAR relational learning scientific knowledge simultaneous localization and mapping social robotics stream mining suport vector machines validation visualization
2017 |
Geilke, Michael; Karwath, Andreas; Frank, Eibe; Kramer, Stefan Online estimation of discrete, continuous, and conditional joint densities using classifier chains Journal Article Data Mining and Knowledge Discovery, pp. 1-43, 2017, ISBN: 1573-756X. Abstract | Links | BibTeX | Tags: density estimation, machine learning, stream mining @article{geilke17, title = {Online estimation of discrete, continuous, and conditional joint densities using classifier chains}, author = {Michael Geilke and Andreas Karwath and Eibe Frank and Stefan Kramer}, url = {https://doi.org/10.1007/s10618-017-0546-6}, doi = {10.1007/s10618-017-0546-6}, isbn = {1573-756X}, year = {2017}, date = {2017-11-17}, journal = {Data Mining and Knowledge Discovery}, pages = {1-43}, publisher = {Springer US}, abstract = {We address the problem of estimating discrete, continuous, and conditional joint densities online, i.e., the algorithm is only provided the current example and its current estimate for its update. The family of proposed online density estimators, estimation of densities online (EDO), uses classifier chains to model dependencies among features, where each classifier in the chain estimates the probability of one particular feature. Because a single chain may not provide a reliable estimate, we also consider ensembles of classifier chains and ensembles of weighted classifier chains. For all density estimators, we provide consistency proofs and propose algorithms to perform certain inference tasks. The empirical evaluation of the estimators is conducted in several experiments and on datasets of up to several millions of instances. In the discrete case, we compare our estimators to density estimates computed by Bayesian structure learners. In the continuous case, we compare them to a state-of-the-art online density estimator. Our experiments demonstrate that, even though designed to work online, EDO delivers estimators of competitive accuracy compared to other density estimators (batch Bayesian structure learners on discrete datasets and the state-of-the-art online density estimator on continuous datasets). Besides achieving similar performance in these cases, EDO is also able to estimate densities with mixed types of variables, i.e., discrete and continuous random variables.}, keywords = {density estimation, machine learning, stream mining}, pubstate = {published}, tppubtype = {article} } We address the problem of estimating discrete, continuous, and conditional joint densities online, i.e., the algorithm is only provided the current example and its current estimate for its update. The family of proposed online density estimators, estimation of densities online (EDO), uses classifier chains to model dependencies among features, where each classifier in the chain estimates the probability of one particular feature. Because a single chain may not provide a reliable estimate, we also consider ensembles of classifier chains and ensembles of weighted classifier chains. For all density estimators, we provide consistency proofs and propose algorithms to perform certain inference tasks. The empirical evaluation of the estimators is conducted in several experiments and on datasets of up to several millions of instances. In the discrete case, we compare our estimators to density estimates computed by Bayesian structure learners. In the continuous case, we compare them to a state-of-the-art online density estimator. Our experiments demonstrate that, even though designed to work online, EDO delivers estimators of competitive accuracy compared to other density estimators (batch Bayesian structure learners on discrete datasets and the state-of-the-art online density estimator on continuous datasets). Besides achieving similar performance in these cases, EDO is also able to estimate densities with mixed types of variables, i.e., discrete and continuous random variables. |
Karwath, Andreas ; Hubrich, Markus ; Kramer, Stefan Artificial Intelligence in Medicine: 16th Conference on Artificial Intelligence in Medicine, AIME 2017, Vienna, Austria, June 21-24, 2017, Proceedings, Springer Springer International Publishing, Cham, 2017, ISBN: 978-3-319-59758-4. Abstract | Links | BibTeX | Tags: alzheimer, deep learning, machine learning, medicine, visualization @conference{karwath2017a, title = {Convolutional Neural Networks for the Identification of Regions of Interests in PET Scans: A Study of Representation Learning for Diagnosing Alzheimer's Disease}, author = {Karwath, Andreas and Hubrich, Markus and Kramer, Stefan}, editor = {en Teije, Annette and Popow, Christian and Holmes, John H. and Sacchi, Lucia}, url = {http://dx.doi.org/10.1007/978-3-319-59758-4_36}, doi = {10.1007/978-3-319-59758-4_36}, isbn = {978-3-319-59758-4}, year = {2017}, date = {2017-06-21}, booktitle = {Artificial Intelligence in Medicine: 16th Conference on Artificial Intelligence in Medicine, AIME 2017, Vienna, Austria, June 21-24, 2017, Proceedings}, pages = {316-321}, publisher = {Springer International Publishing}, address = {Cham}, organization = {Springer}, abstract = {When diagnosing patients suffering from dementia based on imaging data like PET scans, the identification of suitable predictive regions of interest (ROIs) is of great importance. We present a case study of 3-D Convolutional Neural Networks (CNNs) for the detection of ROIs in this context, just using voxel data, without any knowledge given a priori. Our results on data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) suggest that the predictive performance of the method is on par with that of state-of-the-art methods, with the additional benefit of potential insights into affected brain regions.}, keywords = {alzheimer, deep learning, machine learning, medicine, visualization}, pubstate = {published}, tppubtype = {conference} } When diagnosing patients suffering from dementia based on imaging data like PET scans, the identification of suitable predictive regions of interest (ROIs) is of great importance. We present a case study of 3-D Convolutional Neural Networks (CNNs) for the detection of ROIs in this context, just using voxel data, without any knowledge given a priori. Our results on data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) suggest that the predictive performance of the method is on par with that of state-of-the-art methods, with the additional benefit of potential insights into affected brain regions. |
2016 |
Geilke, Michael ; Karwath, Andreas ; Kramer, Stefan Online density estimation of heterogeneous data streams in higher dimensions Conference Machine learning and knowledge discovery in databases : European Conference, ECML PKDD 2016, Riva del Garda, Italy, September 19-23, 2016 : Proceedings Part 1, 2016. Abstract | Links | BibTeX | Tags: data mining, density estimation, stream mining @conference{geilke2016, title = {Online density estimation of heterogeneous data streams in higher dimensions}, author = {Geilke, Michael and Karwath, Andreas and Kramer, Stefan}, url = {http://link.springer.com/chapter/10.1007/978-3-319-46128-1_5}, doi = {doi:10.1007/978-3-319-46128-1_5}, year = {2016}, date = {2016-01-01}, booktitle = {Machine learning and knowledge discovery in databases : European Conference, ECML PKDD 2016, Riva del Garda, Italy, September 19-23, 2016 : Proceedings Part 1}, pages = {65-80}, abstract = {The joint density of a data stream is suitable for performing data mining tasks without having access to the original data. However, the methods proposed so far only target a small to medium number of variables, since their estimates rely on representing all the interdependencies between the variables of the data. High-dimensional data streams, which are becoming more and more frequent due to increasing numbers of interconnected devices, are, therefore, pushing these methods to their limits. To mitigate these limitations, we present an approach that projects the original data stream into a vector space and uses a set of representatives to provide an estimate. Due to the structure of the estimates, it enables the density estimation of higher-dimensional data and approaches the true density with increasing dimensionality of the vector space. Moreover, it is not only designed to estimate homogeneous data, i.e., where all variables are nominal or all variables are numeric, but it can also estimate heterogeneous data. The evaluation is conducted on synthetic and real-world data. The software related to this paper is available at https://github.com/geilke/mideo.}, howpublished = {urlhttps://publications.UB.Uni-Mainz.DE/opus/frontdoor.php?source_opus=54808}, keywords = {data mining, density estimation, stream mining}, pubstate = {published}, tppubtype = {conference} } The joint density of a data stream is suitable for performing data mining tasks without having access to the original data. However, the methods proposed so far only target a small to medium number of variables, since their estimates rely on representing all the interdependencies between the variables of the data. High-dimensional data streams, which are becoming more and more frequent due to increasing numbers of interconnected devices, are, therefore, pushing these methods to their limits. To mitigate these limitations, we present an approach that projects the original data stream into a vector space and uses a set of representatives to provide an estimate. Due to the structure of the estimates, it enables the density estimation of higher-dimensional data and approaches the true density with increasing dimensionality of the vector space. Moreover, it is not only designed to estimate homogeneous data, i.e., where all variables are nominal or all variables are numeric, but it can also estimate heterogeneous data. The evaluation is conducted on synthetic and real-world data. The software related to this paper is available at https://github.com/geilke/mideo. |
2015 |
Geilke, Michael; Karwath, Andreas; Kramer, Stefan Modeling recurrent distributions in streams using possible worlds Conference 2015 IEEE International Conference on Data Science and Advanced Analytics, DSAA 2015, IEEE, 2015, ISBN: 978-1-4673-8272-4. Abstract | Links | BibTeX | Tags: density estimation, machine learning, possible worlds, stream mining @conference{geilke2015, title = {Modeling recurrent distributions in streams using possible worlds}, author = {Michael Geilke and Andreas Karwath and Stefan Kramer}, url = {http://dx.doi.org/10.1109/DSAA.2015.7344814}, doi = {10.1109/DSAA.2015.7344814}, isbn = {978-1-4673-8272-4}, year = {2015}, date = {2015-10-19}, booktitle = {2015 IEEE International Conference on Data Science and Advanced Analytics, DSAA 2015}, pages = {1-9}, publisher = {IEEE}, crossref = {DBLP:conf/dsaa/2015}, abstract = {Discovering changes in the data distribution of streams and discovering recurrent data distributions are challenging problems in data mining and machine learning. Both have received a lot of attention in the context of classification. With the ever increasing growth of data, however, there is a high demand of compact and universal representations of data streams that enable the user to analyze current as well as historic data without having access to the raw data. To make a first step towards this direction, we propose a condensed representation that captures the various - possibly recurrent - data distributions of the stream by extending the notion of possible worlds. The representation enables queries concerning the whole stream and can, hence, serve as a tool for supporting decision-making processes or serve as a basis for implementing data mining and machine learning algorithms on top of it. We evaluate this condensed representation on synthetic and real-world data. }, keywords = {density estimation, machine learning, possible worlds, stream mining}, pubstate = {published}, tppubtype = {conference} } Discovering changes in the data distribution of streams and discovering recurrent data distributions are challenging problems in data mining and machine learning. Both have received a lot of attention in the context of classification. With the ever increasing growth of data, however, there is a high demand of compact and universal representations of data streams that enable the user to analyze current as well as historic data without having access to the raw data. To make a first step towards this direction, we propose a condensed representation that captures the various - possibly recurrent - data distributions of the stream by extending the notion of possible worlds. The representation enables queries concerning the whole stream and can, hence, serve as a tool for supporting decision-making processes or serve as a basis for implementing data mining and machine learning algorithms on top of it. We evaluate this condensed representation on synthetic and real-world data. |
2014 |
Geilke, Michael; Karwath, Andreas; Kramer, Stefan A probabilistic condensed representation of data for stream mining Conference International Conference on Data Science and Advanced Analytics, DSAA 2014, IEEE, 2014. Abstract | Links | BibTeX | Tags: density estimation, machine learning, stream mining @conference{geilke2014, title = {A probabilistic condensed representation of data for stream mining}, author = {Michael Geilke and Andreas Karwath and Stefan Kramer}, url = {http://dx.doi.org/10.1109/DSAA.2014.7058088}, doi = {10.1109/DSAA.2014.7058088}, year = {2014}, date = {2014-10-30}, booktitle = {International Conference on Data Science and Advanced Analytics, DSAA 2014}, pages = {297-303}, publisher = {IEEE}, crossref = {DBLP:conf/dsaa/2014}, abstract = {Data mining and machine learning algorithms usually operate directly on the data. However, if the data is not available at once or consists of billions of instances, these algorithms easily become infeasible with respect to memory and run-time concerns. As a solution to this problem, we propose a framework, called MiDEO (Mining Density Estimates inferred Online), in which algorithms are designed to operate on a condensed representation of the data. In particular, we propose to use density estimates, which are able to represent billions of instances in a compact form and can be updated when new instances arrive. As an example for an algorithm that operates on density estimates, we consider the task of mining association rules, which we consider as a form of simple statements about the data. The algorithm, called POEt (Pattern mining on Online density esTimates), is evaluated on synthetic and real-world data and is compared to state-of-the-art algorithms.}, keywords = {density estimation, machine learning, stream mining}, pubstate = {published}, tppubtype = {conference} } Data mining and machine learning algorithms usually operate directly on the data. However, if the data is not available at once or consists of billions of instances, these algorithms easily become infeasible with respect to memory and run-time concerns. As a solution to this problem, we propose a framework, called MiDEO (Mining Density Estimates inferred Online), in which algorithms are designed to operate on a condensed representation of the data. In particular, we propose to use density estimates, which are able to represent billions of instances in a compact form and can be updated when new instances arrive. As an example for an algorithm that operates on density estimates, we consider the task of mining association rules, which we consider as a form of simple statements about the data. The algorithm, called POEt (Pattern mining on Online density esTimates), is evaluated on synthetic and real-world data and is compared to state-of-the-art algorithms. |
Gütlein, Martin; Karwath, Andreas; Kramer, Stefan CheS-Mapper 2.0 for visual validation of (Q)SAR models Journal Article J. Cheminformatics, 6 (1), pp. 41, 2014. Abstract | Links | BibTeX | Tags: cheminformatics, data mining, graph mining, validation, visualization @article{gutlein2014, title = {CheS-Mapper 2.0 for visual validation of (Q)SAR models}, author = {Martin Gütlein and Andreas Karwath and Stefan Kramer}, url = {http://dx.doi.org/10.1186/s13321-014-0041-7}, doi = {10.1186/s13321-014-0041-7}, year = {2014}, date = {2014-09-23}, journal = {J. Cheminformatics}, volume = {6}, number = {1}, pages = {41}, abstract = {Background Sound statistical validation is important to evaluate and compare the overall performance of (Q)SAR models. However, classical validation does not support the user in better understanding the properties of the model or the underlying data. Even though, a number of visualization tools for analyzing (Q)SAR information in small molecule datasets exist, integrated visualization methods that allow the investigation of model validation results are still lacking. Results We propose visual validation, as an approach for the graphical inspection of (Q)SAR model validation results. The approach applies the 3D viewer CheS-Mapper, an open-source application for the exploration of small molecules in virtual 3D space. The present work describes the new functionalities in CheS-Mapper 2.0, that facilitate the analysis of (Q)SAR information and allows the visual validation of (Q)SAR models. The tool enables the comparison of model predictions to the actual activity in feature space. The approach is generic: It is model-independent and can handle physico-chemical and structural input features as well as quantitative and qualitative endpoints. Conclusions Visual validation with CheS-Mapper enables analyzing (Q)SAR information in the data and indicates how this information is employed by the (Q)SAR model. It reveals, if the endpoint is modeled too specific or too generic and highlights common properties of misclassified compounds. Moreover, the researcher can use CheS-Mapper to inspect how the (Q)SAR model predicts activity cliffs. The CheS-Mapper software is freely available at http://ches-mapper.org. Graphical abstract Comparing actual and predicted activity values with CheS-Mapper.}, keywords = {cheminformatics, data mining, graph mining, validation, visualization}, pubstate = {published}, tppubtype = {article} } Background Sound statistical validation is important to evaluate and compare the overall performance of (Q)SAR models. However, classical validation does not support the user in better understanding the properties of the model or the underlying data. Even though, a number of visualization tools for analyzing (Q)SAR information in small molecule datasets exist, integrated visualization methods that allow the investigation of model validation results are still lacking. Results We propose visual validation, as an approach for the graphical inspection of (Q)SAR model validation results. The approach applies the 3D viewer CheS-Mapper, an open-source application for the exploration of small molecules in virtual 3D space. The present work describes the new functionalities in CheS-Mapper 2.0, that facilitate the analysis of (Q)SAR information and allows the visual validation of (Q)SAR models. The tool enables the comparison of model predictions to the actual activity in feature space. The approach is generic: It is model-independent and can handle physico-chemical and structural input features as well as quantitative and qualitative endpoints. Conclusions Visual validation with CheS-Mapper enables analyzing (Q)SAR information in the data and indicates how this information is employed by the (Q)SAR model. It reveals, if the endpoint is modeled too specific or too generic and highlights common properties of misclassified compounds. Moreover, the researcher can use CheS-Mapper to inspect how the (Q)SAR model predicts activity cliffs. The CheS-Mapper software is freely available at http://ches-mapper.org. Graphical abstract Comparing actual and predicted activity values with CheS-Mapper. |
Seeland, Madeleine; Karwath, Andreas; Kramer, Stefan Structural clustering of millions of molecular graphs Conference Symposium on Applied Computing, SAC 2014, ACM ACM, New York, NY, USA, 2014. Abstract | Links | BibTeX | Tags: @conference{seeland2014a, title = {Structural clustering of millions of molecular graphs}, author = {Madeleine Seeland and Andreas Karwath and Stefan Kramer}, url = {http://doi.acm.org/10.1145/2554850.2555063}, doi = {10.1145/2554850.2555063}, year = {2014}, date = {2014-03-24}, booktitle = {Symposium on Applied Computing, SAC 2014}, pages = {121-128}, publisher = {ACM}, address = {New York, NY, USA}, organization = {ACM}, abstract = {Statistical machine learning algorithms building on patterns found by pattern mining algorithms have to cope with large solution sets and thus the high dimensionality of the feature space. Vice versa, pattern mining algorithms are frequently applied to irrelevant instances, thus causing noise in the output. Solution sets of pattern mining algorithms also typically grow with increasing input datasets. The paper proposes an approach to overcome these limitations. The approach extracts information from trained support vector machines, in particular their support vectors and their relevance according to their coefficients. It uses the support vectors along with their coefficients as input to pattern mining algorithms able to handle weighted instances. Our experiments in the domain of graph mining and molecular graphs show that the resulting models are not significantly less accurate than models trained on the full datasets, yet require only a fraction of the time using much smaller sets of patterns.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } Statistical machine learning algorithms building on patterns found by pattern mining algorithms have to cope with large solution sets and thus the high dimensionality of the feature space. Vice versa, pattern mining algorithms are frequently applied to irrelevant instances, thus causing noise in the output. Solution sets of pattern mining algorithms also typically grow with increasing input datasets. The paper proposes an approach to overcome these limitations. The approach extracts information from trained support vector machines, in particular their support vectors and their relevance according to their coefficients. It uses the support vectors along with their coefficients as input to pattern mining algorithms able to handle weighted instances. Our experiments in the domain of graph mining and molecular graphs show that the resulting models are not significantly less accurate than models trained on the full datasets, yet require only a fraction of the time using much smaller sets of patterns. |
Seeland, Madeleine; Maunz, Andreas; Karwath, Andreas; Kramer, Stefan Extracting information from support vector machines for pattern-based classification Conference Symposium on Applied Computing, SAC 2014, ACM ACM, New York, NY, USA, 2014. Abstract | Links | BibTeX | Tags: @conference{seeland2014b, title = {Extracting information from support vector machines for pattern-based classification}, author = {Madeleine Seeland and Andreas Maunz and Andreas Karwath and Stefan Kramer}, url = {http://dl.acm.org/citation.cfm?doid=2554850.2555065}, doi = {10.1145/2554850.2555065}, year = {2014}, date = {2014-03-24}, booktitle = {Symposium on Applied Computing, SAC 2014}, pages = {129-136}, publisher = {ACM}, address = {New York, NY, USA}, organization = {ACM}, abstract = {Statistical machine learning algorithms building on patterns found by pattern mining algorithms have to cope with large solution sets and thus the high dimensionality of the feature space. Vice versa, pattern mining algorithms are frequently applied to irrelevant instances, thus causing noise in the output. Solution sets of pattern mining algorithms also typically grow with increasing input datasets. The paper proposes an approach to overcome these limitations. The approach extracts information from trained support vector machines, in particular their support vectors and their relevance according to their coefficients. It uses the support vectors along with their coefficients as input to pattern mining algorithms able to handle weighted instances. Our experiments in the domain of graph mining and molecular graphs show that the resulting models are not significantly less accurate than models trained on the full datasets, yet require only a fraction of the time using much smaller sets of patterns.}, keywords = {}, pubstate = {published}, tppubtype = {conference} } Statistical machine learning algorithms building on patterns found by pattern mining algorithms have to cope with large solution sets and thus the high dimensionality of the feature space. Vice versa, pattern mining algorithms are frequently applied to irrelevant instances, thus causing noise in the output. Solution sets of pattern mining algorithms also typically grow with increasing input datasets. The paper proposes an approach to overcome these limitations. The approach extracts information from trained support vector machines, in particular their support vectors and their relevance according to their coefficients. It uses the support vectors along with their coefficients as input to pattern mining algorithms able to handle weighted instances. Our experiments in the domain of graph mining and molecular graphs show that the resulting models are not significantly less accurate than models trained on the full datasets, yet require only a fraction of the time using much smaller sets of patterns. |
2013 |
Geilke, Michael; Frank, Eibe; Karwath, Andreas; Kramer, Stefan Online Estimation of Discrete Densities Conference IEEE 13th International Conference on Data Mining, ICDM 2013, IEEE, 2013, ISSN: 1550-4786. Abstract | Links | BibTeX | Tags: density estimation, machine learning, stream mining @conference{geilke2013, title = {Online Estimation of Discrete Densities}, author = {Michael Geilke and Eibe Frank and Andreas Karwath and Stefan Kramer}, url = {http://dx.doi.org/10.1109/ICDM.2013.91}, doi = {10.1109/ICDM.2013.91}, issn = {1550-4786}, year = {2013}, date = {2013-12-07}, booktitle = {IEEE 13th International Conference on Data Mining, ICDM 2013}, pages = {191-200}, publisher = {IEEE}, crossref = {DBLP:conf/icdm/2013}, abstract = {We address the problem of estimating a discrete joint density online, that is, the algorithm is only provided the current example and its current estimate. The proposed online estimator of discrete densities, EDDO (Estimation of Discrete Densities Online), uses classifier chains to model dependencies among features. Each classifier in the chain estimates the probability of one particular feature. Because a single chain may not provide a reliable estimate, we also consider ensembles of classifier chains and ensembles of weighted classifier chains. For all density estimators, we provide consistency proofs and propose algorithms to perform certain inference tasks. The empirical evaluation of the estimators is conducted in several experiments and on data sets of up to several million instances: We compare them to density estimates computed from Bayesian structure learners, evaluate them under the influence of noise, measure their ability to deal with concept drift, and measure the run-time performance. Our experiments demonstrate that, even though designed to work online, EDDO delivers estimators of competitive accuracy compared to batch Bayesian structure learners and batch variants of EDDO.}, keywords = {density estimation, machine learning, stream mining}, pubstate = {published}, tppubtype = {conference} } We address the problem of estimating a discrete joint density online, that is, the algorithm is only provided the current example and its current estimate. The proposed online estimator of discrete densities, EDDO (Estimation of Discrete Densities Online), uses classifier chains to model dependencies among features. Each classifier in the chain estimates the probability of one particular feature. Because a single chain may not provide a reliable estimate, we also consider ensembles of classifier chains and ensembles of weighted classifier chains. For all density estimators, we provide consistency proofs and propose algorithms to perform certain inference tasks. The empirical evaluation of the estimators is conducted in several experiments and on data sets of up to several million instances: We compare them to density estimates computed from Bayesian structure learners, evaluate them under the influence of noise, measure their ability to deal with concept drift, and measure the run-time performance. Our experiments demonstrate that, even though designed to work online, EDDO delivers estimators of competitive accuracy compared to batch Bayesian structure learners and batch variants of EDDO. |
Gütlein, Martin; Helma, Christoph; Karwath, Andreas; Kramer, Stefan A Large-Scale Empirical Evaluation of Cross-Validation and External Test Set Validation in (Q)SAR Journal Article Molecular Informatics, 32 (5-6), pp. 516-528, 2013. Abstract | Links | BibTeX | Tags: cheminformatics, crossvalidation, external validation, QSAR @article{guetlein2013, title = {A Large-Scale Empirical Evaluation of Cross-Validation and External Test Set Validation in (Q)SAR}, author = {Martin Gütlein and Christoph Helma and Andreas Karwath and Stefan Kramer}, url = {http://onlinelibrary.wiley.com/doi/10.1002/minf.201200134/abstract}, doi = {10.1002/minf.201200134}, year = {2013}, date = {2013-10-14}, journal = {Molecular Informatics}, volume = {32}, number = {5-6}, pages = {516-528}, abstract = {(Q)SAR model validation is essential to ensure the quality of inferred models and to indicate future model predictivity on unseen compounds. Proper validation is also one of the requirements of regulatory authorities in order to accept the (Q)SAR model, and to approve its use in real world scenarios as alternative testing method. However, at the same time, the question of how to validate a (Q)SAR model, in particular whether to employ variants of cross-validation or external test set validation, is still under discussion. In this paper, we empirically compare a k-fold cross-validation with external test set validation. To this end we introduce a workflow allowing to realistically simulate the common problem setting of building predictive models for relatively small datasets. The workflow allows to apply the built and validated models on large amounts of unseen data, and to compare the performance of the different validation approaches. The experimental results indicate that cross-validation produces higher performant (Q)SAR models than external test set validation, reduces the variance of the results, while at the same time underestimates the performance on unseen compounds. The experimental results reported in this paper suggest that, contrary to current conception in the community, cross-validation may play a significant role in evaluating the predictivity of (Q)SAR models.}, keywords = {cheminformatics, crossvalidation, external validation, QSAR}, pubstate = {published}, tppubtype = {article} } (Q)SAR model validation is essential to ensure the quality of inferred models and to indicate future model predictivity on unseen compounds. Proper validation is also one of the requirements of regulatory authorities in order to accept the (Q)SAR model, and to approve its use in real world scenarios as alternative testing method. However, at the same time, the question of how to validate a (Q)SAR model, in particular whether to employ variants of cross-validation or external test set validation, is still under discussion. In this paper, we empirically compare a k-fold cross-validation with external test set validation. To this end we introduce a workflow allowing to realistically simulate the common problem setting of building predictive models for relatively small datasets. The workflow allows to apply the built and validated models on large amounts of unseen data, and to compare the performance of the different validation approaches. The experimental results indicate that cross-validation produces higher performant (Q)SAR models than external test set validation, reduces the variance of the results, while at the same time underestimates the performance on unseen compounds. The experimental results reported in this paper suggest that, contrary to current conception in the community, cross-validation may play a significant role in evaluating the predictivity of (Q)SAR models. |
2012 |
Seeland, Madeleine; Karwath, Andreas; Kramer, Stefan A structural cluster kernel for learning on graphs Conference The 18th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2012, ACM ACM, New York, NY, USA, 2012, ISBN: 978-1-4503-1462-6. Abstract | Links | BibTeX | Tags: cheminformatics, clustering, data mining, kernels, QSAR, suport vector machines @conference{seeland2012, title = {A structural cluster kernel for learning on graphs}, author = {Madeleine Seeland and Andreas Karwath and Stefan Kramer}, url = {http://doi.acm.org/10.1145/2339530.2339614}, doi = {10.1145/2339530.2339614}, isbn = {978-1-4503-1462-6}, year = {2012}, date = {2012-08-12}, booktitle = {The 18th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2012}, pages = {516-524}, publisher = {ACM}, address = {New York, NY, USA}, organization = {ACM}, crossref = {DBLP:conf/kdd/2012}, abstract = {In recent years, graph kernels have received considerable interest within the machine learning and data mining community. Here, we introduce a novel approach enabling kernel methods to utilize additional information hidden in the structural neighborhood of the graphs under consideration. Our novel structural cluster kernel (SCK) incorporates similarities induced by a structural clustering algorithm to improve state-of-the-art graph kernels. The approach taken is based on the idea that graph similarity can not only be described by the similarity between the graphs themselves, but also by the similarity they possess with respect to their structural neighborhood. We applied our novel kernel in a supervised and a semi-supervised setting to regression and classification problems on a number of real-world datasets of molecular graphs. Our results show that the structural cluster similarity information can indeed leverage the prediction performance of the base kernel, particularly when the dataset is structurally sparse and consequently structurally diverse. By additionally taking into account a large number of unlabeled instances the performance of the structural cluster kernel can further be improved.}, keywords = {cheminformatics, clustering, data mining, kernels, QSAR, suport vector machines}, pubstate = {published}, tppubtype = {conference} } In recent years, graph kernels have received considerable interest within the machine learning and data mining community. Here, we introduce a novel approach enabling kernel methods to utilize additional information hidden in the structural neighborhood of the graphs under consideration. Our novel structural cluster kernel (SCK) incorporates similarities induced by a structural clustering algorithm to improve state-of-the-art graph kernels. The approach taken is based on the idea that graph similarity can not only be described by the similarity between the graphs themselves, but also by the similarity they possess with respect to their structural neighborhood. We applied our novel kernel in a supervised and a semi-supervised setting to regression and classification problems on a number of real-world datasets of molecular graphs. Our results show that the structural cluster similarity information can indeed leverage the prediction performance of the base kernel, particularly when the dataset is structurally sparse and consequently structurally diverse. By additionally taking into account a large number of unlabeled instances the performance of the structural cluster kernel can further be improved. |
Gütlein, Martin; Karwath, Andreas; Kramer, Stefan CheS-Mapper - Chemical Space Mapping and Visualization in 3D Journal Article J. Cheminformatics, 4 , pp. 7, 2012. Abstract | Links | BibTeX | Tags: cheminformatics, clustering, dimensionality reduction, QSAR, visualization @article{gutlein2012, title = {CheS-Mapper - Chemical Space Mapping and Visualization in 3D}, author = {Martin Gütlein and Andreas Karwath and Stefan Kramer}, url = {http://dx.doi.org/10.1186/1758-2946-4-7}, doi = {10.1186/1758-2946-4-7}, year = {2012}, date = {2012-03-17}, journal = {J. Cheminformatics}, volume = {4}, pages = {7}, abstract = {Analyzing chemical datasets is a challenging task for scientific researchers in the field of chemoinformatics. It is important, yet difficult to understand the relationship between the structure of chemical compounds, their physico-chemical properties, and biological or toxic effects. To that respect, visualization tools can help to better comprehend the underlying correlations. Our recently developed 3D molecular viewer CheS-Mapper (Chemical Space Mapper) divides large datasets into clusters of similar compounds and consequently arranges them in 3D space, such that their spatial proximity reflects their similarity. The user can indirectly determine similarity, by selecting which features to employ in the process. The tool can use and calculate different kind of features, like structural fragments as well as quantitative chemical descriptors. These features can be highlighted within CheS-Mapper, which aids the chemist to better understand patterns and regularities and relate the observations to established scientific knowledge. As a final function, the tool can also be used to select and export specific subsets of a given dataset for further analysis. }, keywords = {cheminformatics, clustering, dimensionality reduction, QSAR, visualization}, pubstate = {published}, tppubtype = {article} } Analyzing chemical datasets is a challenging task for scientific researchers in the field of chemoinformatics. It is important, yet difficult to understand the relationship between the structure of chemical compounds, their physico-chemical properties, and biological or toxic effects. To that respect, visualization tools can help to better comprehend the underlying correlations. Our recently developed 3D molecular viewer CheS-Mapper (Chemical Space Mapper) divides large datasets into clusters of similar compounds and consequently arranges them in 3D space, such that their spatial proximity reflects their similarity. The user can indirectly determine similarity, by selecting which features to employ in the process. The tool can use and calculate different kind of features, like structural fragments as well as quantitative chemical descriptors. These features can be highlighted within CheS-Mapper, which aids the chemist to better understand patterns and regularities and relate the observations to established scientific knowledge. As a final function, the tool can also be used to select and export specific subsets of a given dataset for further analysis. |
Grzonka, Slawomir; Karwath, Andreas; Dijoux, Frederic; Burgard, Wolfram Activity-Based Estimation of Human Trajectories Journal Article IEEE Transactions on Robotics, 28 (1), pp. 234-245, 2012. Abstract | Links | BibTeX | Tags: activity recognition, machine learning, simultaneous localization and mapping @article{grzonka2012, title = {Activity-Based Estimation of Human Trajectories}, author = {Slawomir Grzonka and Andreas Karwath and Frederic Dijoux and Wolfram Burgard}, url = {http://dx.doi.org/10.1109/TRO.2011.2165372}, doi = {10.1109/TRO.2011.2165372}, year = {2012}, date = {2012-02-02}, journal = {IEEE Transactions on Robotics}, volume = {28}, number = {1}, pages = {234-245}, abstract = {We present a novel approach to incrementally determine the trajectory of a person in 3-D based on its motions and activities in real time. In our algorithm, we estimate the motions and activities of the user given the data that are obtained from a motion capture suit equipped with several inertial measurement units. These activities include walking up and down staircases, as well as opening and closing doors. We interpret the first two types of activities as motion constraints and door-handling events as landmark detections in a graph-based simultaneous localization and mapping (SLAM) framework. Since we cannot distinguish between individual doors, we employ a multihypothesis tracking approach on top of the SLAM procedure to deal with the high data-association uncertainty. As a result, we are able to accurately and robustly recover the trajectory of the person. Additionally, we present an algorithm to build approximate geometrical and topological maps based on the estimated trajectory and detected activities. We evaluate our approach in practical experiments that are carried out with different subjects and in various environments.}, keywords = {activity recognition, machine learning, simultaneous localization and mapping}, pubstate = {published}, tppubtype = {article} } We present a novel approach to incrementally determine the trajectory of a person in 3-D based on its motions and activities in real time. In our algorithm, we estimate the motions and activities of the user given the data that are obtained from a motion capture suit equipped with several inertial measurement units. These activities include walking up and down staircases, as well as opening and closing doors. We interpret the first two types of activities as motion constraints and door-handling events as landmark detections in a graph-based simultaneous localization and mapping (SLAM) framework. Since we cannot distinguish between individual doors, we employ a multihypothesis tracking approach on top of the SLAM procedure to deal with the high data-association uncertainty. As a result, we are able to accurately and robustly recover the trajectory of the person. Additionally, we present an algorithm to build approximate geometrical and topological maps based on the estimated trajectory and detected activities. We evaluate our approach in practical experiments that are carried out with different subjects and in various environments. |
2010 |
Hardy, Barry J; Douglas, Nicki; Helma, Christoph; Rautenberg, Micha; Jeliazkova, Nina; Jeliazkov, Vedrin; Nikolova, Ivelina; Benigni, Romualdo; Tcheremenskaia, Olga; Kramer, Stefan; Girschick, Tobias; Buchwald, Fabian; Wicker, Jörg; Karwath, Andreas; Gütlein, Martin; Maunz, Andreas; Sarimveis, Haralambos; Melagraki, Georgia; Afantitis, Antreas; Sopasakis, Pantelis; Gallagher, David; Poroikov, Vladimir; Filimonov, Dmitry; Zakharov, Alexey V; Lagunin, Alexey; Gloriozova, Tatyana; Novikov, Sergey; Skvortsova, Natalia; Druzhilovsky, Dmitry; Chawla, Sunil; Ghosh, Indira; Ray, Surajit; Patel, Hitesh; Escher, Sylvia Collaborative development of predictive toxicology applications Journal Article J. Cheminformatics, 2 , pp. 7, 2010. Abstract | Links | BibTeX | Tags: crossvalidation, data mining, QSAR, scientific knowledge @article{hardy2010, title = {Collaborative development of predictive toxicology applications}, author = {Barry J. Hardy and Nicki Douglas and Christoph Helma and Micha Rautenberg and Nina Jeliazkova and Vedrin Jeliazkov and Ivelina Nikolova and Romualdo Benigni and Olga Tcheremenskaia and Stefan Kramer and Tobias Girschick and Fabian Buchwald and Jörg Wicker and Andreas Karwath and Martin Gütlein and Andreas Maunz and Haralambos Sarimveis and Georgia Melagraki and Antreas Afantitis and Pantelis Sopasakis and David Gallagher and Vladimir Poroikov and Dmitry Filimonov and Alexey V. Zakharov and Alexey Lagunin and Tatyana Gloriozova and Sergey Novikov and Natalia Skvortsova and Dmitry Druzhilovsky and Sunil Chawla and Indira Ghosh and Surajit Ray and Hitesh Patel and Sylvia Escher}, url = {http://dx.doi.org/10.1186/1758-2946-2-7}, doi = {10.1186/1758-2946-2-7}, year = {2010}, date = {2010-08-31}, journal = {J. Cheminformatics}, volume = {2}, pages = {7}, abstract = {OpenTox provides an interoperable, standards-based Framework for the support of predictive toxicology data management, algorithms, modelling, validation and reporting. It is relevant to satisfying the chemical safety assessment requirements of the REACH legislation as it supports access to experimental data, (Quantitative) Structure-Activity Relationship models, and toxicological information through an integrating platform that adheres to regulatory requirements and OECD validation principles. Initial research defined the essential components of the Framework including the approach to data access, schema and management, use of controlled vocabularies and ontologies, architecture, web service and communications protocols, and selection and integration of algorithms for predictive modelling. OpenTox provides end-user oriented tools to non-computational specialists, risk assessors, and toxicological experts in addition to Application Programming Interfaces (APIs) for developers of new applications. OpenTox actively supports public standards for data representation, interfaces, vocabularies and ontologies, Open Source approaches to core platform components, and community-based collaboration approaches, so as to progress system interoperability goals. The OpenTox Framework includes APIs and services for compounds, datasets, features, algorithms, models, ontologies, tasks, validation, and reporting which may be combined into multiple applications satisfying a variety of different user needs. OpenTox applications are based on a set of distributed, interoperable OpenTox API-compliant REST web services. The OpenTox approach to ontology allows for efficient mapping of complementary data coming from different datasets into a unifying structure having a shared terminology and representation. Two initial OpenTox applications are presented as an illustration of the potential impact of OpenTox for high-quality and consistent structure-activity relationship modelling of REACH-relevant endpoints: ToxPredict which predicts and reports on toxicities for endpoints for an input chemical structure, and ToxCreate which builds and validates a predictive toxicity model based on an input toxicology dataset. Because of the extensible nature of the standardised Framework design, barriers of interoperability between applications and content are removed, as the user may combine data, models and validation from multiple sources in a dependable and time-effective way.}, keywords = {crossvalidation, data mining, QSAR, scientific knowledge}, pubstate = {published}, tppubtype = {article} } OpenTox provides an interoperable, standards-based Framework for the support of predictive toxicology data management, algorithms, modelling, validation and reporting. It is relevant to satisfying the chemical safety assessment requirements of the REACH legislation as it supports access to experimental data, (Quantitative) Structure-Activity Relationship models, and toxicological information through an integrating platform that adheres to regulatory requirements and OECD validation principles. Initial research defined the essential components of the Framework including the approach to data access, schema and management, use of controlled vocabularies and ontologies, architecture, web service and communications protocols, and selection and integration of algorithms for predictive modelling. OpenTox provides end-user oriented tools to non-computational specialists, risk assessors, and toxicological experts in addition to Application Programming Interfaces (APIs) for developers of new applications. OpenTox actively supports public standards for data representation, interfaces, vocabularies and ontologies, Open Source approaches to core platform components, and community-based collaboration approaches, so as to progress system interoperability goals. The OpenTox Framework includes APIs and services for compounds, datasets, features, algorithms, models, ontologies, tasks, validation, and reporting which may be combined into multiple applications satisfying a variety of different user needs. OpenTox applications are based on a set of distributed, interoperable OpenTox API-compliant REST web services. The OpenTox approach to ontology allows for efficient mapping of complementary data coming from different datasets into a unifying structure having a shared terminology and representation. Two initial OpenTox applications are presented as an illustration of the potential impact of OpenTox for high-quality and consistent structure-activity relationship modelling of REACH-relevant endpoints: ToxPredict which predicts and reports on toxicities for endpoints for an input chemical structure, and ToxCreate which builds and validates a predictive toxicity model based on an input toxicology dataset. Because of the extensible nature of the standardised Framework design, barriers of interoperability between applications and content are removed, as the user may combine data, models and validation from multiple sources in a dependable and time-effective way. |
Grzonka, Slawomir; Dijoux, Frederic; Karwath, Andreas; Burgard, Wolfram Mapping indoor environments based on human activity Conference IEEE International Conference on Robotics and Automation, ICRA 2010, IEEE, 2010, ISBN: 978-1-4244-5038-1. Abstract | Links | BibTeX | Tags: activity recognition, machine learning, simultaneous localization and mapping @conference{grzonka2010b, title = {Mapping indoor environments based on human activity}, author = {Slawomir Grzonka and Frederic Dijoux and Andreas Karwath and Wolfram Burgard}, url = {http://dx.doi.org/10.1109/ROBOT.2010.5509976}, doi = {10.1109/ROBOT.2010.5509976}, isbn = {978-1-4244-5038-1}, year = {2010}, date = {2010-05-03}, booktitle = {IEEE International Conference on Robotics and Automation, ICRA 2010}, pages = {476-481}, publisher = {IEEE}, crossref = {DBLP:conf/icra/2010}, abstract = {We present a novel approach to build approximate maps of structured environments utilizing human motion and activity. Our approach uses data recorded with a data suit which is equipped with several IMUs to detect movements of a person and door opening and closing events. In our approach we interpret the movements as motion constraints and door handling events as landmark detections in a graph-based SLAM framework. As we cannot distinguish between individual doors, we employ a multi-hypothesis approach on top of the SLAM system to deal with the high data-association uncertainty. As a result, our approach is able to accurately and robustly recover the trajectory of the person. We additionally take advantage of the fact that people traverse free space and that doors separate rooms to recover the geometric structure of the environment after the graph optimization. We evaluate our approach in several experiments carried out with different users and in environments of different types. }, keywords = {activity recognition, machine learning, simultaneous localization and mapping}, pubstate = {published}, tppubtype = {conference} } We present a novel approach to build approximate maps of structured environments utilizing human motion and activity. Our approach uses data recorded with a data suit which is equipped with several IMUs to detect movements of a person and door opening and closing events. In our approach we interpret the movements as motion constraints and door handling events as landmark detections in a graph-based SLAM framework. As we cannot distinguish between individual doors, we employ a multi-hypothesis approach on top of the SLAM system to deal with the high data-association uncertainty. As a result, our approach is able to accurately and robustly recover the trajectory of the person. We additionally take advantage of the fact that people traverse free space and that doors separate rooms to recover the geometric structure of the environment after the graph optimization. We evaluate our approach in several experiments carried out with different users and in environments of different types. |
Grzonka, Slawomir; Dijoux, Frederic; Karwath, Andreas; Burgard, Wolfram Learning Maps of Indoor Environments Based on Human Activity Conference Embedded Reasoning, Papers from the 2010 AAAI Spring Symposium, 2010. Abstract | Links | BibTeX | Tags: activity recognition, localization, machine learning, mobile systems and mobility, simultaneous localization and mapping, social robotics @conference{grzonka2010a, title = {Learning Maps of Indoor Environments Based on Human Activity}, author = {Slawomir Grzonka and Frederic Dijoux and Andreas Karwath and Wolfram Burgard}, url = {http://www.aaai.org/ocs/index.php/SSS/SSS10/paper/view/1172}, year = {2010}, date = {2010-03-23}, booktitle = {Embedded Reasoning, Papers from the 2010 AAAI Spring Symposium}, crossref = {DBLP:conf/aaaiss/2010-4}, abstract = {We present a novel approach to build approximate maps of structured environments utilizing human motion and activity. Our approach uses data recorded with a data suit which is equipped with several IMUs to detect movements of a person and door opening and closing events. In our approach we interpret the movements as motion constraints and door handling events as landmark detections in a graph-based SLAM framework. As we cannot distinguish between individual doors, we employ a multi-hypothesis approach on top of the SLAM system to deal with the high data-association uncertainty. As a result, our approach is able to accurately and robustly recover the trajectory of the person. We additionally take advantage of the fact that people traverse free space and that doors separate rooms to recover the geometric structure of the environment after the graph optimization. We evaluate our approach in several experiments carried out with different users and in environments of different types.}, keywords = {activity recognition, localization, machine learning, mobile systems and mobility, simultaneous localization and mapping, social robotics}, pubstate = {published}, tppubtype = {conference} } We present a novel approach to build approximate maps of structured environments utilizing human motion and activity. Our approach uses data recorded with a data suit which is equipped with several IMUs to detect movements of a person and door opening and closing events. In our approach we interpret the movements as motion constraints and door handling events as landmark detections in a graph-based SLAM framework. As we cannot distinguish between individual doors, we employ a multi-hypothesis approach on top of the SLAM system to deal with the high data-association uncertainty. As a result, our approach is able to accurately and robustly recover the trajectory of the person. We additionally take advantage of the fact that people traverse free space and that doors separate rooms to recover the geometric structure of the environment after the graph optimization. We evaluate our approach in several experiments carried out with different users and in environments of different types. |
2009 |
Gütlein, Martin; Frank, Eibe; Hall, Mark A; Karwath, Andreas Large-scale attribute selection using wrappers Conference The IEEE Symposium on Computational Intelligence and Data Mining, CIDM 2009, IEEE, 2009, ISBN: 978-1-4244-2765-9. Abstract | Links | BibTeX | Tags: crossvalidation, machine learning @conference{gutlein2009, title = {Large-scale attribute selection using wrappers}, author = {Martin Gütlein and Eibe Frank and Mark A. Hall and Andreas Karwath}, url = {http://dx.doi.org/10.1109/CIDM.2009.4938668}, doi = {10.1109/CIDM.2009.4938668}, isbn = {978-1-4244-2765-9}, year = {2009}, date = {2009-01-01}, booktitle = {The IEEE Symposium on Computational Intelligence and Data Mining, CIDM 2009}, pages = {332-339}, publisher = {IEEE}, crossref = {DBLP:conf/cidm/2009}, abstract = {Scheme-specific attribute selection with the wrapper and variants of forward selection is a popular attribute selection technique for classification that yields good results. However, it can run the risk of overfitting because of the extent of the search and the extensive use of internal cross-validation. Moreover, although wrapper evaluators tend to achieve superior accuracy compared to filters, they face a high computational cost. The problems of overfitting and high runtime occur in particular on high-dimensional datasets, like microarray data. We investigate Linear Forward Selection, a technique to reduce the number of attributes expansions in each forward selection step. Our experiments demonstrate that this approach is faster, finds smaller subsets and can even increase the accuracy compared to standard forward selection. We also investigate a variant that applies explicit subset size determination in forward selection to combat overfitting, where the search is forced to stop at a precomputed ldquooptimalrdquo subset size. We show that this technique reduces subset size while maintaining comparable accuracy.}, keywords = {crossvalidation, machine learning}, pubstate = {published}, tppubtype = {conference} } Scheme-specific attribute selection with the wrapper and variants of forward selection is a popular attribute selection technique for classification that yields good results. However, it can run the risk of overfitting because of the extent of the search and the extensive use of internal cross-validation. Moreover, although wrapper evaluators tend to achieve superior accuracy compared to filters, they face a high computational cost. The problems of overfitting and high runtime occur in particular on high-dimensional datasets, like microarray data. We investigate Linear Forward Selection, a technique to reduce the number of attributes expansions in each forward selection step. Our experiments demonstrate that this approach is faster, finds smaller subsets and can even increase the accuracy compared to standard forward selection. We also investigate a variant that applies explicit subset size determination in forward selection to combat overfitting, where the search is forced to stop at a precomputed ldquooptimalrdquo subset size. We show that this technique reduces subset size while maintaining comparable accuracy. |
Schulz, Hannes; Kersting, Kristian; Karwath, Andreas ILP, the Blind, and the Elephant: Euclidean Embedding of Co-proven Queries Conference Inductive Logic Programming, 19th International Conference, ILP 2009, Springer-Verlag Berlin Heidelberg Springer Verlag, Berlin Heidelberg, Germany, 2009, ISBN: 978-3-642-13839-3. Abstract | Links | BibTeX | Tags: cheminformatics, dimensionality reduction, inductive logic programming, relational learning, scientific knowledge, visualization @conference{schulz2009, title = {ILP, the Blind, and the Elephant: Euclidean Embedding of Co-proven Queries}, author = {Hannes Schulz and Kristian Kersting and Andreas Karwath}, url = {http://dx.doi.org/10.1007/978-3-642-13840-9_20}, doi = {10.1007/978-3-642-13840-9_20}, isbn = {978-3-642-13839-3}, year = {2009}, date = {2009-01-01}, booktitle = {Inductive Logic Programming, 19th International Conference, ILP 2009}, pages = {209-216}, publisher = {Springer Verlag}, address = {Berlin Heidelberg, Germany}, organization = {Springer-Verlag Berlin Heidelberg}, crossref = {DBLP:conf/ilp/2009}, abstract = {Relational data is complex. This complexity makes one of the basic steps of ILP difficult: understanding the data and results. If the user cannot easily understand it, he draws incomplete conclusions. The situation is very much as in the parable of the blind men and the elephant that appears in many cultures. In this tale the blind work independently and with quite different pieces of information, thereby drawing very different conclusions about the nature of the beast. In contrast, visual representations make it easy to shift from one perspective to another while exploring and analyzing data. This paper describes a method for embedding interpretations and queries into a single, common Euclidean space based on their co-proven statistics. We demonstrate our method on real-world datasets showing that ILP results can indeed be captured at a glance.}, keywords = {cheminformatics, dimensionality reduction, inductive logic programming, relational learning, scientific knowledge, visualization}, pubstate = {published}, tppubtype = {conference} } Relational data is complex. This complexity makes one of the basic steps of ILP difficult: understanding the data and results. If the user cannot easily understand it, he draws incomplete conclusions. The situation is very much as in the parable of the blind men and the elephant that appears in many cultures. In this tale the blind work independently and with quite different pieces of information, thereby drawing very different conclusions about the nature of the beast. In contrast, visual representations make it easy to shift from one perspective to another while exploring and analyzing data. This paper describes a method for embedding interpretations and queries into a single, common Euclidean space based on their co-proven statistics. We demonstrate our method on real-world datasets showing that ILP results can indeed be captured at a glance. |
2008 |
Karwath, Andreas; Kersting, Kristian; Landwehr, Niels Boosting Relational Sequence Alignments Conference The 8th IEEE International Conference on Data Mining, ICDM 2008, IEEE, 2008, ISBN: 978-0-7695-3502-9. Abstract | Links | BibTeX | Tags: inductive logic programming, machine learning, relational learning, scientific knowledge @conference{karwath2008, title = {Boosting Relational Sequence Alignments}, author = {Andreas Karwath and Kristian Kersting and Niels Landwehr}, url = {http://dx.doi.org/10.1109/ICDM.2008.127}, doi = {10.1109/ICDM.2008.127}, isbn = {978-0-7695-3502-9}, year = {2008}, date = {2008-12-15}, booktitle = {The 8th IEEE International Conference on Data Mining, ICDM 2008}, pages = {857-862}, publisher = {IEEE}, crossref = {DBLP:conf/icdm/2008}, abstract = {The task of aligning sequences arises in many applications. Classical dynamic programming approaches require the explicit state enumeration in the reward model. This is often impractical: the number of states grows very quickly with the number of domain objects and relations among these objects. Relational sequence alignment aims at exploiting symbolic structure to avoid the full enumeration. This comes at the expense of a more complex reward model selection problem: virtually infinitely many abstraction levels have to be explored. In this paper, we apply gradient-based boosting to leverage this problem. Specifically, we show how to reduce the learning problem to a series of relational regressions problems. The main benefit of this is that interactions between states variables are introduced only as needed, so that the potentially infinite search space is not explicitly considered. As our experimental results show, this boosting approach can significantly improve upon established results in challenging applications.}, keywords = {inductive logic programming, machine learning, relational learning, scientific knowledge}, pubstate = {published}, tppubtype = {conference} } The task of aligning sequences arises in many applications. Classical dynamic programming approaches require the explicit state enumeration in the reward model. This is often impractical: the number of states grows very quickly with the number of domain objects and relations among these objects. Relational sequence alignment aims at exploiting symbolic structure to avoid the full enumeration. This comes at the expense of a more complex reward model selection problem: virtually infinitely many abstraction levels have to be explored. In this paper, we apply gradient-based boosting to leverage this problem. Specifically, we show how to reduce the learning problem to a series of relational regressions problems. The main benefit of this is that interactions between states variables are introduced only as needed, so that the potentially infinite search space is not explicitly considered. As our experimental results show, this boosting approach can significantly improve upon established results in challenging applications. |
Kersting, Kristian; De Raedt, Luc ; Gutmann, Bernd; Karwath, Andreas; Landwehr, Niels Relational Sequence Learning Book Chapter Probabilistic Inductive Logic Programming - Theory and Applications, 4911 , pp. 28-55, Springer Verlag, Berlin Heidelberg, Germany, 2008, ISBN: 978-3-540-78651-1. Abstract | Links | BibTeX | Tags: inductive logic programming, machine learning, relational learning, scientific knowledge @inbook{kersting2008, title = {Relational Sequence Learning}, author = {Kristian Kersting and De Raedt, Luc and Bernd Gutmann and Andreas Karwath and Niels Landwehr}, url = {http://dx.doi.org/10.1007/978-3-540-78652-8_2}, doi = {10.1007/978-3-540-78652-8_2}, isbn = {978-3-540-78651-1}, year = {2008}, date = {2008-01-01}, booktitle = {Probabilistic Inductive Logic Programming - Theory and Applications}, volume = {4911}, pages = {28-55}, publisher = {Springer Verlag}, address = {Berlin Heidelberg, Germany}, organization = {Springer-Verlag Berlin Heidelberg}, crossref = {DBLP:conf/ilp/2008p}, abstract = {Sequential behavior and sequence learning are essential to intelligence. Often the elements of sequences exhibit an internal structure that can elegantly be represented using relational atoms. Applying traditional sequential learning techniques to such relational sequences requires one either to ignore the internal structure or to live with a combinatorial explosion of the model complexity. This chapter briefly reviews relational sequence learning and describes several techniques tailored towards realizing this, such as local pattern mining techniques, (hidden) Markov models, conditional random fields, dynamic programming and reinforcement learning.}, keywords = {inductive logic programming, machine learning, relational learning, scientific knowledge}, pubstate = {published}, tppubtype = {inbook} } Sequential behavior and sequence learning are essential to intelligence. Often the elements of sequences exhibit an internal structure that can elegantly be represented using relational atoms. Applying traditional sequential learning techniques to such relational sequences requires one either to ignore the internal structure or to live with a combinatorial explosion of the model complexity. This chapter briefly reviews relational sequence learning and describes several techniques tailored towards realizing this, such as local pattern mining techniques, (hidden) Markov models, conditional random fields, dynamic programming and reinforcement learning. |
2007 |
King, Ross D; Karwath, Andreas; Clare, Amanda; Dehaspe, Luc Logic and the Automatic Acquisition of Scientific Knowledge: An Application to Functional Genomics Conference Computational Discovery of Scientific Knowledge, Introduction, Techniques, and Applications in Environmental and Life Sciences, 4660 , Lecture Notes in Computer Science Springer-Verlag Berlin Heidelberg Springer Verlag, Berlin Heidelberg, Germany, 2007, ISBN: 978-3-540-73919-7. Abstract | Links | BibTeX | Tags: bioinformatics, data mining, inductive logic programming, machine learning, relational learning, scientific knowledge @conference{king2007, title = {Logic and the Automatic Acquisition of Scientific Knowledge: An Application to Functional Genomics}, author = {Ross D. King and Andreas Karwath and Amanda Clare and Luc Dehaspe}, url = {http://dx.doi.org/10.1007/978-3-540-73920-3_13}, doi = {10.1007/978-3-540-73920-3_13}, isbn = {978-3-540-73919-7}, year = {2007}, date = {2007-01-01}, booktitle = {Computational Discovery of Scientific Knowledge, Introduction, Techniques, and Applications in Environmental and Life Sciences}, volume = {4660}, pages = {273-289}, publisher = {Springer Verlag}, address = {Berlin Heidelberg, Germany}, organization = {Springer-Verlag Berlin Heidelberg}, series = {Lecture Notes in Computer Science}, crossref = {DBLP:conf/dis/2007book}, abstract = {This paper is a manifesto aimed at computer scientists interested in developing and applying scientific discovery methods. It argues that: science is experiencing an unprecedented “explosion” in the amount of available data; traditional data analysis methods cannot deal with this increased quantity of data; there is an urgent need to automate the process of refining scientific data into scientific knowledge; inductive logic programming (ILP) is a data analysis framework well suited for this task; and exciting new scientific discoveries can be achieved using ILP scientific discovery methods. We describe an example of using ILP to analyse a large and complex bioinformatic database that has produced unexpected and interesting scientific results in functional genomics. We then point a possible way forward to integrating machine learning with scientific databases to form intelligent databases.}, keywords = {bioinformatics, data mining, inductive logic programming, machine learning, relational learning, scientific knowledge}, pubstate = {published}, tppubtype = {conference} } This paper is a manifesto aimed at computer scientists interested in developing and applying scientific discovery methods. It argues that: science is experiencing an unprecedented “explosion” in the amount of available data; traditional data analysis methods cannot deal with this increased quantity of data; there is an urgent need to automate the process of refining scientific data into scientific knowledge; inductive logic programming (ILP) is a data analysis framework well suited for this task; and exciting new scientific discoveries can be achieved using ILP scientific discovery methods. We describe an example of using ILP to analyse a large and complex bioinformatic database that has produced unexpected and interesting scientific results in functional genomics. We then point a possible way forward to integrating machine learning with scientific databases to form intelligent databases. |
Karwath, Andreas; Kersting, Kristian Relational Sequence Alignments and Logos Conference Inductive Logic Programming, 16th International Conference, ILP 2006, 4455 , Lecture Notes in Computer Science Springer-Verlag Berlin Heidelberg Springer Verlag, Berlin Heidelberg, Germany, 2007, ISBN: 978-3-540-73846-6. Abstract | Links | BibTeX | Tags: bioinformatics, inductive logic programming, relational learning, scientific knowledge @conference{karwath2007, title = {Relational Sequence Alignments and Logos}, author = {Andreas Karwath and Kristian Kersting}, url = {http://dx.doi.org/10.1007/978-3-540-73847-3_29}, doi = {10.1007/978-3-540-73847-3_29}, isbn = {978-3-540-73846-6}, year = {2007}, date = {2007-01-01}, booktitle = {Inductive Logic Programming, 16th International Conference, ILP 2006}, volume = {4455}, pages = {290-304}, publisher = {Springer Verlag}, address = {Berlin Heidelberg, Germany}, organization = {Springer-Verlag Berlin Heidelberg}, series = {Lecture Notes in Computer Science}, crossref = {DBLP:conf/ilp/2006}, abstract = {The need to measure sequence similarity arises in many applicitation domains and often coincides with sequence alignment: the more similar two sequences are, the better they can be aligned. Aligning sequences not only shows how similar sequences are, it also shows where there are differences and correspondences between the sequences. Traditionally, the alignment has been considered for sequences of flat symbols only. Many real world sequences such as natural language sentences and protein secondary structures, however, exhibit rich internal structures. This is akin to the problem of dealing with structured examples studied in the field of inductive logic programming (ILP). In this paper, we introduce Real, which is a powerful, yet simple approach to align sequence of structured symbols using well-established ILP distance measures within traditional alignment methods. Although straight-forward, experiments on protein data and Medline abstracts show that this approach works well in practice, that the resulting alignments can indeed provide more information than flat ones, and that they are meaningful to experts when represented graphically.}, keywords = {bioinformatics, inductive logic programming, relational learning, scientific knowledge}, pubstate = {published}, tppubtype = {conference} } The need to measure sequence similarity arises in many applicitation domains and often coincides with sequence alignment: the more similar two sequences are, the better they can be aligned. Aligning sequences not only shows how similar sequences are, it also shows where there are differences and correspondences between the sequences. Traditionally, the alignment has been considered for sequences of flat symbols only. Many real world sequences such as natural language sentences and protein secondary structures, however, exhibit rich internal structures. This is akin to the problem of dealing with structured examples studied in the field of inductive logic programming (ILP). In this paper, we introduce Real, which is a powerful, yet simple approach to align sequence of structured symbols using well-established ILP distance measures within traditional alignment methods. Although straight-forward, experiments on protein data and Medline abstracts show that this approach works well in practice, that the resulting alignments can indeed provide more information than flat ones, and that they are meaningful to experts when represented graphically. |
2006 |
Karwath, Andreas; De Raedt, Luc SMIREP: Predicting Chemical Activity from SMILES Journal Article Journal of Chemical Information and Modeling, 46 (6), pp. 2432 - 2444, 2006. Abstract | Links | BibTeX | Tags: cheminformatics, graph mining, machine learning, QSAR, relational learning, scientific knowledge @article{karwath06c, title = {SMIREP: Predicting Chemical Activity from SMILES}, author = {Andreas Karwath and De Raedt, Luc}, url = {http://pubs.acs.org/doi/abs/10.1021/ci060159g}, doi = {10.1021/ci060159g}, year = {2006}, date = {2006-10-12}, journal = {Journal of Chemical Information and Modeling}, volume = {46}, number = {6}, pages = {2432 - 2444}, abstract = {Most approaches to structure-activity-relationship (SAR) prediction proceed in two steps. In the first step, a typically large set of fingerprints, or fragments of interest, is constructed (either by hand or by some recent data mining techniques). In the second step, machine learning techniques are applied to obtain a predictive model. The result is often not only a highly accurate but also hard to interpret model. In this paper, we demonstrate the capabilities of a novel SAR algorithm, SMIREP, which tightly integrates the fragment and model generation steps and which yields simple models in the form of a small set of IF-THEN rules. These rules contain SMILES fragments, which are easy to understand to the computational chemist. SMIREP combines ideas from the well-known IREP rule learner with a novel fragmentation algorithm for SMILES strings. SMIREP has been evaluated on three problems: the prediction of binding activities for the estrogen receptor (Environmental Protection Agency's (EPA's) Distributed Structure-Searchable Toxicity (DSSTox) National Center for Toxicological Research estrogen receptor (NCTRER) Database), the prediction of mutagenicity using the carcinogenic potency database (CPDB), and the prediction of biodegradability on a subset of the Environmental Fate Database (EFDB). In these applications, SMIREP has the advantage of producing easily interpretable rules while having predictive accuracies that are comparable to those of alternative state-of-the-art techniques.}, keywords = {cheminformatics, graph mining, machine learning, QSAR, relational learning, scientific knowledge}, pubstate = {published}, tppubtype = {article} } Most approaches to structure-activity-relationship (SAR) prediction proceed in two steps. In the first step, a typically large set of fingerprints, or fragments of interest, is constructed (either by hand or by some recent data mining techniques). In the second step, machine learning techniques are applied to obtain a predictive model. The result is often not only a highly accurate but also hard to interpret model. In this paper, we demonstrate the capabilities of a novel SAR algorithm, SMIREP, which tightly integrates the fragment and model generation steps and which yields simple models in the form of a small set of IF-THEN rules. These rules contain SMILES fragments, which are easy to understand to the computational chemist. SMIREP combines ideas from the well-known IREP rule learner with a novel fragmentation algorithm for SMILES strings. SMIREP has been evaluated on three problems: the prediction of binding activities for the estrogen receptor (Environmental Protection Agency's (EPA's) Distributed Structure-Searchable Toxicity (DSSTox) National Center for Toxicological Research estrogen receptor (NCTRER) Database), the prediction of mutagenicity using the carcinogenic potency database (CPDB), and the prediction of biodegradability on a subset of the Environmental Fate Database (EFDB). In these applications, SMIREP has the advantage of producing easily interpretable rules while having predictive accuracies that are comparable to those of alternative state-of-the-art techniques. |
Karwath, Andreas; Kersting, Kristian Relational Sequence Alignments Conference Proc. The 4th International Workshop on Mining and Learning with Graphs, MLG 2006, % editor = Thomas Gärtner and Gemma C. Garriga and Thorsten Meinl, % month = September, 2006, (workshop). BibTeX | Tags: bioinformatics, cheminformatics, relational learning, scientific knowledge @conference{karwath06b, title = {Relational Sequence Alignments}, author = {Andreas Karwath and Kristian Kersting}, year = {2006}, date = {2006-01-01}, booktitle = {Proc. The 4th International Workshop on Mining and Learning with Graphs, MLG 2006, % editor = Thomas Gärtner and Gemma C. Garriga and Thorsten Meinl, % month = September}, pages = {149-156}, note = {workshop}, keywords = {bioinformatics, cheminformatics, relational learning, scientific knowledge}, pubstate = {published}, tppubtype = {conference} } |
Clare, Amanda; Karwath, Andreas; Ougham, Helen; King, Ross D Functional bioinformatics for Arabidopsis thaliana Journal Article Bioinformatics, 22 (9), pp. 1130-1136, 2006. Abstract | Links | BibTeX | Tags: bioinformatics, data mining, inductive logic programming, machine learning, relational learning, scientific knowledge @article{karwath06a, title = {Functional bioinformatics for Arabidopsis thaliana}, author = {Amanda Clare and Andreas Karwath and Helen Ougham and Ross D. King}, url = {https://bioinformatics.oxfordjournals.org/content/22/9/1130.full.pdf+html}, doi = {10.1093/bioinformatics/btl051}, year = {2006}, date = {2006-01-01}, journal = {Bioinformatics}, volume = {22}, number = {9}, pages = {1130-1136}, abstract = {Motivation: The genome of Arabidopsis thaliana, which has the best understood plant genome, still has approximately one-third of its genes with no functional annotation at all from either MIPS or TAIR. We have applied our Data Mining Prediction (DMP) method to the problem of predicting the functional classes of these protein sequences. This method is based on using a hybrid machine-learning/data-mining method to identify patterns in the bioinformatic data about sequences that are predictive of function. We use data about sequence, predicted secondary structure, predicted structural domain, InterPro patterns, sequence similarity profile and expressions data. Results: We predicted the functional class of a high percentage of the Arabidopsis genes with currently unknown function. These predictions are interpretable and have good test accuracies. We describe in detail seven of the rules produced. Availability: Rulesets are available at http://www.aber.ac.uk/compsci/Research/bio/dss/arabpreds/ and predictions are available at http://www.genepredictions.org Contact:afc@aber.ac.uk}, keywords = {bioinformatics, data mining, inductive logic programming, machine learning, relational learning, scientific knowledge}, pubstate = {published}, tppubtype = {article} } Motivation: The genome of Arabidopsis thaliana, which has the best understood plant genome, still has approximately one-third of its genes with no functional annotation at all from either MIPS or TAIR. We have applied our Data Mining Prediction (DMP) method to the problem of predicting the functional classes of these protein sequences. This method is based on using a hybrid machine-learning/data-mining method to identify patterns in the bioinformatic data about sequences that are predictive of function. We use data about sequence, predicted secondary structure, predicted structural domain, InterPro patterns, sequence similarity profile and expressions data. Results: We predicted the functional class of a high percentage of the Arabidopsis genes with currently unknown function. These predictions are interpretable and have good test accuracies. We describe in detail seven of the rules produced. Availability: Rulesets are available at http://www.aber.ac.uk/compsci/Research/bio/dss/arabpreds/ and predictions are available at http://www.genepredictions.org Contact:afc@aber.ac.uk |
Backofen, Rolf; Borrmann, Hans-Gunther; Deck, Werner; Dedner, Andreas; De Raedt, Luc ; Desch, Klaus; Diesmann, Markus; Geier, Martin; Greiner, Andreas; Hess, Wolfgang R; Honerkamp, Josef; Jankowski, Stefan; Krossing, Ingo; Liehr, Andreas W; Karwath, Andreas; Klöfkorn, Robert; Pesché, Raphaël; Potjans, Tobias C; Röttger, Michael C; Schmidt-Thieme, Lars; Schneider, Gerhard; Voß, Björn; Wiebelt, Bernd; Wienemann, Peter; Winterer, Volker-Henning A Bottom-up approach to Grid-Computing at a University: the Black-Forest-Grid Initiative Journal Article Praxis der Informationsverarbeitung und Kommunikation, 29 (2), pp. 81-87, 2006. Abstract | Links | BibTeX | Tags: data mining, HPC @article{backofen2006, title = {A Bottom-up approach to Grid-Computing at a University: the Black-Forest-Grid Initiative}, author = {Rolf Backofen and Hans-Gunther Borrmann and Werner Deck and Andreas Dedner and De Raedt, Luc and Klaus Desch and Markus Diesmann and Martin Geier and Andreas Greiner and Wolfgang R. Hess and Josef Honerkamp and Stefan Jankowski and Ingo Krossing and Andreas W. Liehr and Andreas Karwath and Robert Klöfkorn and Raphaël Pesché and Tobias C. Potjans and Michael C. Röttger and Lars Schmidt-Thieme and Gerhard Schneider and Björn Voß and Bernd Wiebelt and Peter Wienemann and Volker-Henning Winterer}, url = {http://dx.doi.org/10.1515/PIKO.2006.81}, doi = {10.1515/PIKO.2006.81}, year = {2006}, date = {2006-01-01}, journal = {Praxis der Informationsverarbeitung und Kommunikation}, volume = {29}, number = {2}, pages = {81-87}, abstract = {Recent years have seen a rapid increase in the need for highperformance computing. These demands come from disciplines such as particle physics traditionally relying on High Performance Computing (HPC) but lately also from the various branches of life science that have matured into quantitative disciplines. The classical infrastructure of university computer centres results to be unsuited to cope with the new requirements for a multitude of reasons. Here we discuss the causes of this failure and present a solution developed at the University of Freiburg in a collaborative effort of several faculties. We demonstrate that using state of the art grid computing technology the problem can now be addressed in a bottom-up approach. The organizational, technical, and financial components of our framework, the Black Forest Grid Initiative (BFG) are described and results of its implementation are presented. In the process, a number of new questions have emerged which the next phase of our project needs to address.}, keywords = {data mining, HPC}, pubstate = {published}, tppubtype = {article} } Recent years have seen a rapid increase in the need for highperformance computing. These demands come from disciplines such as particle physics traditionally relying on High Performance Computing (HPC) but lately also from the various branches of life science that have matured into quantitative disciplines. The classical infrastructure of university computer centres results to be unsuited to cope with the new requirements for a multitude of reasons. Here we discuss the causes of this failure and present a solution developed at the University of Freiburg in a collaborative effort of several faculties. We demonstrate that using state of the art grid computing technology the problem can now be addressed in a bottom-up approach. The organizational, technical, and financial components of our framework, the Black Forest Grid Initiative (BFG) are described and results of its implementation are presented. In the process, a number of new questions have emerged which the next phase of our project needs to address. |
2005 |
Stolle, Christian; Karwath, Andreas; De Raedt, Luc CLASSIC'CL: an integrated ILP system Conference Proc. 8th International Conference of Discovery Science, DS 2005, 3735 , Lecture Notes in Artificial Intelligence Springer, 2005, ISBN: 978-3-540-29230-2, (Conference). Abstract | Links | BibTeX | Tags: data mining, machine learning, relational learning @conference{karwath05a, title = {CLASSIC'CL: an integrated ILP system}, author = {Christian Stolle and Andreas Karwath and De Raedt, Luc}, url = {http://link.springer.com/chapter/10.1007%2F11563983_31}, doi = {10.1007/11563983_31}, isbn = {978-3-540-29230-2}, year = {2005}, date = {2005-10-08}, booktitle = {Proc. 8th International Conference of Discovery Science, DS 2005}, volume = {3735}, pages = {354-362}, publisher = {Springer}, series = {Lecture Notes in Artificial Intelligence}, abstract = {A novel inductive logic programming system, called Classic’cl is presented. Classic’cl integrates several settings for learning, in particular learning from interpretations and learning from satisfiability. Within these settings, it addresses descriptive and probabilistic modeling tasks. As such, Classic’cl (C-armr, cLAudien, icl-S(S)at, ICl, and CLlpad) integrates several well-known inductive logic programming systems such as Claudien, Warmr (and its extension C-armr), ICL, ICL-SAT, and LLPAD. We report on the implementation, the integration issues as well as on some experiments that compare Classic’cl with some of its predecessors.}, note = {Conference}, keywords = {data mining, machine learning, relational learning}, pubstate = {published}, tppubtype = {conference} } A novel inductive logic programming system, called Classic’cl is presented. Classic’cl integrates several settings for learning, in particular learning from interpretations and learning from satisfiability. Within these settings, it addresses descriptive and probabilistic modeling tasks. As such, Classic’cl (C-armr, cLAudien, icl-S(S)at, ICl, and CLlpad) integrates several well-known inductive logic programming systems such as Claudien, Warmr (and its extension C-armr), ICL, ICL-SAT, and LLPAD. We report on the implementation, the integration issues as well as on some experiments that compare Classic’cl with some of its predecessors. |
2004 |
Bringmann, Björn; Karwath, Andreas Frequent SMILES Miscellaneous Lernen, Wissensentdeckung und Adaptivität, Workshop GI Fachgruppe Maschinelles Lernen, part of LWA, 2004, (Berlin, Germany). Abstract | BibTeX | Tags: cheminformatics, graph mining, machine learning @misc{wshp-fgml-BringmannK04, title = {Frequent SMILES}, author = {Björn Bringmann and Andreas Karwath}, year = {2004}, date = {2004-10-01}, abstract = {Predictive graph mining approaches in chemical databases are extremely popular and effective. Most of these approaches first extract frequent sub-graphs and then use them as features to build predictive models. In the work presented here, the approach taken is similar. However, instead of frequent sub-graphs, frequent trees, based on SMILES strings are derived. For this, the SMILES strings of chemical compounds are decomposed into fragment trees, which in turn are mined for interesting sub-trees. These tree based patterns are then used as features by a classifier to build predictive models. The approach is experimentally evaluated on a real world chemical data set.}, howpublished = {Lernen, Wissensentdeckung und Adaptivität, Workshop GI Fachgruppe Maschinelles Lernen, part of LWA}, note = {Berlin, Germany}, keywords = {cheminformatics, graph mining, machine learning}, pubstate = {published}, tppubtype = {misc} } Predictive graph mining approaches in chemical databases are extremely popular and effective. Most of these approaches first extract frequent sub-graphs and then use them as features to build predictive models. In the work presented here, the approach taken is similar. However, instead of frequent sub-graphs, frequent trees, based on SMILES strings are derived. For this, the SMILES strings of chemical compounds are decomposed into fragment trees, which in turn are mined for interesting sub-trees. These tree based patterns are then used as features by a classifier to build predictive models. The approach is experimentally evaluated on a real world chemical data set. |
Karwath, Andreas; De Raedt, Luc Predictive Graph Mining Conference The International Workshop on Mining Graphs, Trees and Sequences, MGTS 2004, 2004, (workshop). BibTeX | Tags: cheminformatics, graph mining, machine learning, QSAR @conference{karwath04b, title = {Predictive Graph Mining}, author = {Andreas Karwath and De Raedt, Luc}, year = {2004}, date = {2004-09-01}, booktitle = {The International Workshop on Mining Graphs, Trees and Sequences, MGTS 2004}, pages = {25-36}, note = {workshop}, keywords = {cheminformatics, graph mining, machine learning, QSAR}, pubstate = {published}, tppubtype = {conference} } |
Karwath, Andreas; De Raedt, Luc Predictive Graph Mining Conference The 7th International Conference of Discovery Science, DS 2004, 3245 , Lecture Notes in Artificial Intelligence Springer-Verlag Berlin Heidelberg Springer Verlag, Berlin Heidelberg, Germany, 2004, ISBN: 978-3-540-23357-2. Abstract | Links | BibTeX | Tags: cheminformatics, graph mining, machine learning, QSAR @conference{karwath04a, title = {Predictive Graph Mining}, author = {Andreas Karwath and De Raedt, Luc}, url = {http://link.springer.com/chapter/10.1007%2F978-3-540-30214-8_1}, doi = {10.1007/978-3-540-30214-8_1}, isbn = {978-3-540-23357-2}, year = {2004}, date = {2004-01-01}, booktitle = {The 7th International Conference of Discovery Science, DS 2004}, volume = {3245}, pages = {1-15}, publisher = {Springer Verlag}, address = {Berlin Heidelberg, Germany}, organization = {Springer-Verlag Berlin Heidelberg}, series = {Lecture Notes in Artificial Intelligence}, abstract = {Graph mining approaches are extremely popular and effective in molecular databases. The vast majority of these approaches first derive interesting, i.e. frequent, patterns and then use these as features to build predictive models. Rather than building these models in a two step indirect way, the SMIREP system introduced in this paper, derives predictive rule models from molecular data directly. SMIREP combines the SMILES and SMARTS representation languages that are popular in computational chemistry with the IREP rule-learning algorithm by Fürnkranz. Even though SMIREP is focused on SMILES, its principles are also applicable to graph mining problems in other domains. SMIREP is experimentally evaluated on two benchmark databases.}, keywords = {cheminformatics, graph mining, machine learning, QSAR}, pubstate = {published}, tppubtype = {conference} } Graph mining approaches are extremely popular and effective in molecular databases. The vast majority of these approaches first derive interesting, i.e. frequent, patterns and then use these as features to build predictive models. Rather than building these models in a two step indirect way, the SMIREP system introduced in this paper, derives predictive rule models from molecular data directly. SMIREP combines the SMILES and SMARTS representation languages that are popular in computational chemistry with the IREP rule-learning algorithm by Fürnkranz. Even though SMIREP is focused on SMILES, its principles are also applicable to graph mining problems in other domains. SMIREP is experimentally evaluated on two benchmark databases. |
2002 |
Karwath, Andreas; King, Ross D Homology Induction: the use of machine learning to improve sequence similarity searches Journal Article BMC Bioinformatics, 3 (1), 2002. Abstract | Links | BibTeX | Tags: bioinformatics, data mining, inductive logic programming, machine learning, relational learning @article{karwath02a, title = {Homology Induction: the use of machine learning to improve sequence similarity searches}, author = {Andreas Karwath and Ross D. King}, url = {http://bmcbioinformatics.biomedcentral.com/articles/10.1186/1471-2105-3-11}, doi = {10.1186/1471-2105-3-11}, year = {2002}, date = {2002-04-23}, journal = {BMC Bioinformatics}, volume = {3}, number = {1}, abstract = {Background The inference of homology between proteins is a key problem in molecular biology The current best approaches only identify ~50% of homologies (with a false positive rate set at 1/1000). Results We present Homology Induction (HI), a new approach to inferring homology. HI uses machine learning to bootstrap from standard sequence similarity search methods. First a standard method is run, then HI learns rules which are true for sequences of high similarity to the target (assumed homologues) and not true for general sequences, these rules are then used to discriminate sequences in the twilight zone. To learn the rules HI describes the sequences in a novel way based on a bioinformatic knowledge base, and the machine learning method of inductive logic programming. To evaluate HI we used the PDB40D benchmark which lists sequences of known homology but low sequence similarity. We compared the HI methodoly with PSI-BLAST alone and found HI performed significantly better. In addition, Receiver Operating Characteristic (ROC) curve analysis showed that these improvements were robust for all reasonable error costs. The predictive homology rules learnt by HI by can be interpreted biologically to provide insight into conserved features of homologous protein families. Conclusions HI is a new technique for the detection of remote protein homolgy – a central bioinformatic problem. HI with PSI-BLAST is shown to outperform PSI-BLAST for all error costs. It is expect that similar improvements would be obtained using HI with any sequence similarity method. }, keywords = {bioinformatics, data mining, inductive logic programming, machine learning, relational learning}, pubstate = {published}, tppubtype = {article} } Background The inference of homology between proteins is a key problem in molecular biology The current best approaches only identify ~50% of homologies (with a false positive rate set at 1/1000). Results We present Homology Induction (HI), a new approach to inferring homology. HI uses machine learning to bootstrap from standard sequence similarity search methods. First a standard method is run, then HI learns rules which are true for sequences of high similarity to the target (assumed homologues) and not true for general sequences, these rules are then used to discriminate sequences in the twilight zone. To learn the rules HI describes the sequences in a novel way based on a bioinformatic knowledge base, and the machine learning method of inductive logic programming. To evaluate HI we used the PDB40D benchmark which lists sequences of known homology but low sequence similarity. We compared the HI methodoly with PSI-BLAST alone and found HI performed significantly better. In addition, Receiver Operating Characteristic (ROC) curve analysis showed that these improvements were robust for all reasonable error costs. The predictive homology rules learnt by HI by can be interpreted biologically to provide insight into conserved features of homologous protein families. Conclusions HI is a new technique for the detection of remote protein homolgy – a central bioinformatic problem. HI with PSI-BLAST is shown to outperform PSI-BLAST for all error costs. It is expect that similar improvements would be obtained using HI with any sequence similarity method. |
Karwath, Andreas Large Logical Đatabases and their Applications to Molecular Biology PhD Thesis University of Wales, Aberystwyth, 2002. BibTeX | Tags: bioinformatics, data mining, inductive logic programming, machine learning, relational learning, scientific knowledge @phdthesis{karwath02b, title = {Large Logical Đatabases and their Applications to Molecular Biology}, author = {Andreas Karwath}, year = {2002}, date = {2002-01-01}, school = {University of Wales, Aberystwyth}, keywords = {bioinformatics, data mining, inductive logic programming, machine learning, relational learning, scientific knowledge}, pubstate = {published}, tppubtype = {phdthesis} } |
2001 |
Karwath, Andreas; King, Ross D An automated ILP server in the field of bioinformatics Conference The Eleventh International Conference on Inductive Logic Programming, ILP 2001, 2157 , Lecture Notes in Computer Science Springer-Verlag Berlin Heidelberg Springer Verlag, Berlin Heidelberg, Germany, 2001, ISBN: 978-3-540-42538-0. Abstract | Links | BibTeX | Tags: bioinformatics, data mining, inductive logic programming, machine learning, relational learning @conference{Karwath2001, title = {An automated ILP server in the field of bioinformatics}, author = {Andreas Karwath and Ross D. King}, editor = {Raghu Ramakrishnan and Michele Sebag}, url = {http://link.springer.com/chapter/10.1007%2F3-540-44797-0_8}, doi = {10.1007/3-540-44797-0_8}, isbn = {978-3-540-42538-0}, year = {2001}, date = {2001-09-09}, booktitle = {The Eleventh International Conference on Inductive Logic Programming, ILP 2001}, volume = {2157}, pages = {91-103}, publisher = {Springer Verlag}, address = {Berlin Heidelberg, Germany}, organization = {Springer-Verlag Berlin Heidelberg}, series = {Lecture Notes in Computer Science}, abstract = {The identification of evolutionary related (homologous) proteins is a key problem in molecular biology. Here we present a inductive logic programming based method, Homology Induction (HI), which acts as a filter for existing sequence similarity searches to improve their performance in the detection of remote protein homologies. HI performs a PSI-BLAST search to generate positive, negative, and uncertain examples, and collects descriptions of these examples. It then learns rules to discriminate the positive and negative examples. The rules are used to filter the uncertain examples in the “twilight zone”. HI uses a multitable database of 51,430,710 pre-fabricated facts from a variety of biological sources, and the inductive logic programming system Aleph to induce rules. Hi was tested on an independent set of protein sequences with equal or less than 40 per cent sequence similarity (PDB40D). ROC analysis is performed showing that HI can significantly improve existing similarity searches. The method is automated and can be used via a web/mail interface.}, keywords = {bioinformatics, data mining, inductive logic programming, machine learning, relational learning}, pubstate = {published}, tppubtype = {conference} } The identification of evolutionary related (homologous) proteins is a key problem in molecular biology. Here we present a inductive logic programming based method, Homology Induction (HI), which acts as a filter for existing sequence similarity searches to improve their performance in the detection of remote protein homologies. HI performs a PSI-BLAST search to generate positive, negative, and uncertain examples, and collects descriptions of these examples. It then learns rules to discriminate the positive and negative examples. The rules are used to filter the uncertain examples in the “twilight zone”. HI uses a multitable database of 51,430,710 pre-fabricated facts from a variety of biological sources, and the inductive logic programming system Aleph to induce rules. Hi was tested on an independent set of protein sequences with equal or less than 40 per cent sequence similarity (PDB40D). ROC analysis is performed showing that HI can significantly improve existing similarity searches. The method is automated and can be used via a web/mail interface. |
King, Ross D; Karwath, Andreas; Clare, Amanda; Dehaspe, Luc The utility of different representations of protein sequence for predicting functional class Journal Article Bioinformatics, 17 (5), pp. 445-454, 2001. Abstract | Links | BibTeX | Tags: bioinformatics, data mining, inductive logic programming, relational learning, scientific knowledge @article{King2001a, title = {The utility of different representations of protein sequence for predicting functional class}, author = {Ross D. King and Andreas Karwath and Amanda Clare and Luc Dehaspe}, url = {https://bioinformatics.oxfordjournals.org/content/17/5/445}, doi = {10.1093/bioinformatics/17.5.445}, year = {2001}, date = {2001-01-19}, journal = {Bioinformatics}, volume = {17}, number = {5}, pages = {445-454}, abstract = {Motivation: Data Mining Prediction (DMP) is a novel approach to predicting protein functional class from sequence. DMP works even in the absence of a homologous protein of known function. We investigate the utility of different ways of representing protein sequence in DMP (residue frequencies, phylogeny, predicted structure) using the Escherichia coli genome as a model. Results: Using the different representations DMP learnt prediction rules that were more accurate than default at every level of function using every type of representation. The most effective way to represent sequence was using phylogeny (75% accuracy and 13% coverage of unassigned ORFs at the most general level of function: 69% accuracy and 7% coverage at the most detailed). We tested different methods for combining predictions from the different types of representation. These improved both the accuracy and coverage of predictions, e.g. 40% of all unassigned ORFs could be predicted at an estimated accuracy of 60% and 5% of unassigned ORFs could be predicted at an estimated accuracy of 86%. Availability: The rules and data are freely available. Warmr is free to academics. Contact: rdk@aber.ac.uk}, keywords = {bioinformatics, data mining, inductive logic programming, relational learning, scientific knowledge}, pubstate = {published}, tppubtype = {article} } Motivation: Data Mining Prediction (DMP) is a novel approach to predicting protein functional class from sequence. DMP works even in the absence of a homologous protein of known function. We investigate the utility of different ways of representing protein sequence in DMP (residue frequencies, phylogeny, predicted structure) using the Escherichia coli genome as a model. Results: Using the different representations DMP learnt prediction rules that were more accurate than default at every level of function using every type of representation. The most effective way to represent sequence was using phylogeny (75% accuracy and 13% coverage of unassigned ORFs at the most general level of function: 69% accuracy and 7% coverage at the most detailed). We tested different methods for combining predictions from the different types of representation. These improved both the accuracy and coverage of predictions, e.g. 40% of all unassigned ORFs could be predicted at an estimated accuracy of 60% and 5% of unassigned ORFs could be predicted at an estimated accuracy of 86%. Availability: The rules and data are freely available. Warmr is free to academics. Contact: rdk@aber.ac.uk |
2000 |
King, Ross D; Karwath, Andreas; Clare, Amanda; Dehaspe, Luc Accurate prediction of protein functional class from sequence in the Mycobacterium tuberculosis and Escherichia coli genomes using data mining. Journal Article Yeast (Comparative and Functional Genomics), 17 , pp. 283-293, 2000. Abstract | Links | BibTeX | Tags: bioinformatics, data mining, inductive logic programming, relational learning, scientific knowledge @article{king00a, title = {Accurate prediction of protein functional class from sequence in the Mycobacterium tuberculosis and Escherichia coli genomes using data mining.}, author = {Ross D. King and Andreas Karwath and Amanda Clare and Luc Dehaspe}, url = {http://onlinelibrary.wiley.com/doi/10.1002/1097-0061(200012)17:4%3C283::AID-YEA52%3E3.0.CO;2-F/abstract}, doi = {10.1002/1097-0061(200012)17:4<283::AID-YEA52>3.0.CO;2-F}, year = {2000}, date = {2000-12-08}, journal = {Yeast (Comparative and Functional Genomics)}, volume = {17}, pages = {283-293}, abstract = {The analysis of genomics data needs to become as automated as its generation. Here we present a novel data-mining approach to predicting protein functional class from sequence. This method is based on a combination of inductive logic programming clustering and rule learning. We demonstrate the effectiveness of this approach on the M. tuberculosis and E. coli genomes, and identify biologically interpretable rules which predict protein functional class from information only available from the sequence. These rules predict 65% of the ORFs with no assigned function in M. tuberculosis and 24% of those in E. coli, with an estimated accuracy of 60–80% (depending on the level of functional assignment). The rules are founded on a combination of detection of remote homology, convergent evolution and horizontal gene transfer. We identify rules that predict protein functional class even in the absence of detectable sequence or structural homology. These rules give insight into the evolutionary history of M. tuberculosis and E. coli. }, keywords = {bioinformatics, data mining, inductive logic programming, relational learning, scientific knowledge}, pubstate = {published}, tppubtype = {article} } The analysis of genomics data needs to become as automated as its generation. Here we present a novel data-mining approach to predicting protein functional class from sequence. This method is based on a combination of inductive logic programming clustering and rule learning. We demonstrate the effectiveness of this approach on the M. tuberculosis and E. coli genomes, and identify biologically interpretable rules which predict protein functional class from information only available from the sequence. These rules predict 65% of the ORFs with no assigned function in M. tuberculosis and 24% of those in E. coli, with an estimated accuracy of 60–80% (depending on the level of functional assignment). The rules are founded on a combination of detection of remote homology, convergent evolution and horizontal gene transfer. We identify rules that predict protein functional class even in the absence of detectable sequence or structural homology. These rules give insight into the evolutionary history of M. tuberculosis and E. coli. |
King, Ross D; Karwath, Andreas; Clare, Amanda; Dehaspe, Luc Logic and the Automatic Acquisition of Scientific Knowledge Journal Article EACIS (Electronic Articles in Computer and Information Science), 5 (031), 2000. Abstract | Links | BibTeX | Tags: bioinformatics, data mining, scientific knowledge @article{King2000c, title = {Logic and the Automatic Acquisition of Scientific Knowledge}, author = {Ross D. King and Andreas Karwath and Amanda Clare and Luc Dehaspe}, url = {http://www.ida.liu.se/ext/epa/cis/mi-17/02/orig.html}, year = {2000}, date = {2000-12-01}, journal = {EACIS (Electronic Articles in Computer and Information Science)}, volume = {5}, number = {031}, abstract = {This paper is a manifesto. It argues that: Science is experiencing an unprecedented "explosion" in the amount of available data. Traditional data analysis methods cannot deal with this increased quantity of data. There is therefore an urgent need to automate the process of refining scientific data into scientific knowledge. Inductive logic programming (ILP) is the data analysis framework best suited for this task. We describe an example of using ILP to analyse a large and complex bioinformatic database which produced unexpected and interesting scientific results. We then point a possible way forward to integrating machine learning with scientific databases to form intelligent inductive databases.}, keywords = {bioinformatics, data mining, scientific knowledge}, pubstate = {published}, tppubtype = {article} } This paper is a manifesto. It argues that: Science is experiencing an unprecedented "explosion" in the amount of available data. Traditional data analysis methods cannot deal with this increased quantity of data. There is therefore an urgent need to automate the process of refining scientific data into scientific knowledge. Inductive logic programming (ILP) is the data analysis framework best suited for this task. We describe an example of using ILP to analyse a large and complex bioinformatic database which produced unexpected and interesting scientific results. We then point a possible way forward to integrating machine learning with scientific databases to form intelligent inductive databases. |
King, Ross D; Karwath, Andreas; Clare, Amanda; Dehaspe, Luc Genome scale prediction of protein functional class from sequence using data mining Conference The Sixth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2000, The Association for Computing Machinery, New York, USA, 2000. Links | BibTeX | Tags: bioinformatics, data mining, inductive logic programming, relational learning @conference{King2000, title = {Genome scale prediction of protein functional class from sequence using data mining}, author = {Ross D. King and Andreas Karwath and Amanda Clare and Luc Dehaspe}, editor = {Raghu Ramakrishnan and S. Stolfo and R. Bayardo and I. Parsa}, url = {http://doi.acm.org/10.1145/347090.347172}, doi = {10.1145/347090.347172}, year = {2000}, date = {2000-01-01}, booktitle = {The Sixth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2000}, pages = {384-389}, publisher = {The Association for Computing Machinery, New York, USA}, keywords = {bioinformatics, data mining, inductive logic programming, relational learning}, pubstate = {published}, tppubtype = {conference} } |