Extracting Knowledge from High-Throughput Screening Data: Towards the Generation of Biophore Models

Here we present the novel software tool HTSview, designed for the extraction of knowledge from HTS data. The tool combines cheminformatics concepts with data mining and visualization methods towards the identification of appropriate chemical series for optimization. Biophore models are generated that capture relevant motifs from an HTS experiment. In this context, we define a biophore as the fragment-based description of an ensemble of molecules with similar biological activities and structures, based on the feature tree descriptor (M. Rarey and J.S. Dixon, JCAMD, (12), 1998). Feature trees are fragment-based, non-linear, topological descriptors. They describe molecules by a tree structure representing their major chemical building blocks and their connection. Each node corresponds to a small molecular fragment.
The interactive software tool HTSview applies this descriptor to the analysis of HTS data. First, a training set is selected from actives and similar, but inactive compounds. Their feature trees are iteratively aligned onto the largest tree which is used as a reference. This results in a topological template with an invariant core and variable regions. The model consists of a mapping of all feature trees so that the most similar fragments are aligned. From these multiple alignments, biophore models are generated by using statistical methods to identify those features correlated with biological activity. The procedure results in the generation of models, which reveal important structure-activity relationships. Initial results indicate that the approach can be applied to the interpretation of HTS data.