An infrastructure to mine molecular descriptors for ligand selection on virtual screening.

The receptor-ligand interaction evaluation is one important step in rational drug design. The databases that provide the structures of the ligands are growing on a daily basis. This makes it impossible to test all the ligands for a target receptor. Hence, a ligand selection before testing the ligands is needed. One possible approach is to evaluate a set of molecular descriptors. With the aim of describing the characteristics of promising compounds for a specific receptor we introduce a data warehouse-based infrastructure to mine molecular descriptors for virtual screening (VS). We performed experiments that consider as target the receptor HIV-1 protease and different compounds for this protein. A set of 9 molecular descriptors are taken as the predictive attributes and the free energy of binding is taken as a target attribute. By applying the J48 algorithm over the data we obtain decision tree models that achieved up to 84% of accuracy. The models indicate which molecular descriptors and their respective values are relevant to influence good FEB results. Using their rules we performed ligand selection on ZINC database. Our results show important reduction in ligands selection to be applied in VS experiments; for instance, the best selection model picked only 0.21% of the total amount of drug-like ligands.

Automatic design of decision-tree induction algorithms tailored to flexible-receptor docking data.

BACKGROUND: This paper addresses the prediction of the free energy of binding of a drug candidate with enzyme InhA associated with Mycobacterium tuberculosis. This problem is found within rational drug design, where interactions between drug candidates and target proteins are verified through molecular docking simulations. In this application, it is important not only to correctly predict the free energy of binding, but also to provide a comprehensible model that could be validated by a domain specialist. Decision-tree induction algorithms have been successfully used in drug-design related applications, specially considering that decision trees are simple to understand, interpret, and validate. There are several decision-tree induction algorithms available for general-use, but each one has a bias that makes it more suitable for a particular data distribution. In this article, we propose and investigate the automatic design of decision-tree induction algorithms tailored to particular drug-enzyme binding data sets. We investigate the performance of our new method for evaluating binding conformations of different drug candidates to InhA, and we analyze our findings with respect to decision tree accuracy, comprehensibility, and biological relevance. RESULTS: The empirical analysis indicates that our method is capable of automatically generating decision-tree induction algorithms that significantly outperform the traditional C4.5 algorithm with respect to both accuracy and comprehensibility. In addition, we provide the biological interpretation of the rules generated by our approach, reinforcing the importance of comprehensible predictive models in this particular bioinformatics application. CONCLUSIONS: We conclude that automatically designing a decision-tree algorithm tailored to molecular docking data is a promising alternative for the prediction of the free energy from the binding of a drug candidate with a flexible-receptor.

Mining flexible-receptor docking experiments to select promising protein receptor snapshots.

BACKGROUND: Molecular docking simulation is the Rational Drug Design (RDD) step that investigates the affinity between protein receptors and ligands. Typically, molecular docking algorithms consider receptors as rigid bodies. Receptors are, however, intrinsically flexible in the cellular environment. The use of a time series of receptor conformations is an approach to explore its flexibility in molecular docking computer simulations, but it is extensively time-consuming. Hence, selection of the most promising conformations can accelerate docking experiments and, consequently, the RDD efforts. RESULTS: We previously docked four ligands (NADH, TCL, PIF and ETH) to 3,100 conformations of the InhA receptor from M. tuberculosis. Based on the receptor residues-ligand distances we preprocessed all docking results to generate appropriate input to mine data. Data preprocessing was done by calculating the shortest interatomic distances between the ligand and the receptor's residues for each docking result. They were the predictive attributes. The target attribute was the estimated free-energy of binding (FEB) value calculated by the AutodDock3.0.5 software. The mining inputs were submitted to the M5P model tree algorithm. It resulted in short and understandable trees. On the basis of the correlation values, for NADH, TCL and PIF we obtained more than 95% correlation while for ETH, only about 60%. Post processing the generated model trees for each of its linear models (LMs), we calculated the average FEB for their associated instances. From these values we considered a LM as representative if its average FEB was smaller than or equal the average FEB of the test set. The instances in the selected LMs were considered the most promising snapshots. It totalized 1,521, 1,780, 2,085 and 902 snapshots, for NADH, TCL, PIF and ETH respectively. CONCLUSIONS: By post processing the generated model trees we were able to propose a criterion of selection of linear models which, in turn, is capable of selecting a set of promising receptor conformations. As future work we intend to go further and use these results to elaborate a strategy to preprocess the receptors 3-D spatial conformation in order to predict FEB values. Besides, we intend to select other compounds, among the million catalogued, that may be promising as new drug candidates for our particular protein receptor target.