Supervised machine learning techniques to predict binding affinity. A study for cyclin-dependent kinase 2.

Here we report the development of a machine-learning model to predict binding affinity based on the crystallographic structures of protein-ligand complexes. We used an ensemble of crystallographic structures (resolution better than 1.5 Å resolution) for which half-maximal inhibitory concentration (IC50) data is available. Polynomial scoring functions were built using as explanatory variables the energy terms present in the MolDock and PLANTS scoring functions. Prediction performance was tested and the supervised machine learning models showed improvement in the prediction power, when compared with PLANTS and MolDock scoring functions. In addition, the machine-learning model was applied to predict binding affinity of CDK2, which showed a better performance when compared with AutoDock4, AutoDock Vina, MolDock, and PLANTS scores.

SAnDReS a Computational Tool for Statistical Analysis of Docking Results and Development of Scoring Functions.

BACKGROUND: Docking allows to predict ligand binding to proteins, since the 3D-structure for the target is available. Several docking studies have been carried out to identify potential ligands for drug targets. Many of these studies resulted in the leads that were later developed as drugs. OBJECTIVE: Our goal here is to describe the development of an integrated computational tool to assess docking accuracy and build new scoring functions to predict ligandbinding affinity. METHOD: We carried out docking simulations using MVD program for a data set available on CSAR 2014 database (coagulation factor Xa) for which ligand-binding information and structures are available. These docking results were analyzed using SAnDReS available at www.sandres.net. Machine learning methods were applied to build new scoring functions and our results were compared with previously published benchmarks. RESULTS: Our integrated docking strategy generated poses with docking accuracy higher than previously published benchmarks. In addition, the new scoring function developed using SAnDReS shows better performance than well-established scoring functions such the ones available in Autodock, Autodock- Vina, Gold, Glide, and MVD. CONCLUSION: The big data generated during docking lacked an integrated computational tool for statistical analysis of the influence of structural parameters on docking and scoring function performance. Here we describe methods to evaluate docking results using SAnDReS, a computational environment for statistical analysis of docking results and development of scoring functions. We believe that SAnDReS is a computational tool with potential to improve accuracy in docking projects.