Neural Information Processing

Applications to Problems in Bio- and Chemoinformatics

Many data sets in bio- and chemoinformatics are good examples for "structured data", i.e. for data for which a vectorial representation is not recommended or even wrong. Here we develop learning algorithms for classification, regression, and exploratory data analysis for these domains. These algorithms are based on recent approaches to learning on structured representations which have been pursued in the machine learning community. Data sets from the bio- and chemoinformatics domains also serve as a testbed for methods we have developed in the past years (cf. Learning on Structured Representations). Current applications include the classification of DNA microarray data, the analysis of protein sequences, and the analysis of protein sequences, and the analysis of quantitative structure activity relationships (QSAR) for chemical compounds. Datasets and analysis problems also serve as a testbed for algorithms which we have developed in our machine learning projects (cf. "Research" page Learning on Structured Representations).

Acknowledgement: This work was funded by the Technische Universität Berlin.