Automatic generation of bioinformatics tools for predicting protein–ligand binding sites

Motivation: Predictive tools that model protein–ligand binding on demand are needed to promote ligand research in an innovative drug-design environment. However, it takes considerable time and effort to develop predictive tools that can be applied to individual ligands. An automated production pipeline that can rapidly and efficiently develop user-friendly protein–ligand binding predictive tools would be useful. Results: We developed a system for automatically generating protein–ligand binding predictions. Implementation of this system in a pipeline of Semantic Web technique-based web tools will allow users to specify a ligand and receive the tool within 0.5–1 day. We demonstrated high prediction accuracy for three machine learning algorithms and eight ligands. Availability and implementation: The source code and web application are freely available for download at http://utprot.net. They are implemented in Python and supported on Linux. Contact: shimizu@bi.a.u-tokyo.ac.jp Supplementary information: Supplementary data are available at Bioinformatics online.