A Bayesian Target Predictor Method based on Molecular Pairing Energies estimation

Abstract

Virtual screening (VS) is applied in the early drug discovery phases for the quick inspection of huge molecular databases to identify those compounds that most likely bind to a given drug target. In this context, there is the necessity of the use of compact molecular models for database screening and precise target prediction in reasonable times. In this work we present a new compact energy-based model that is tested for its application to Virtual Screening and target prediction. The model can be used to quickly identify active compounds in huge databases based on the estimation of the molecule's pairing energies. The greatest molecular polar regions along with its geometrical distribution are considered by using a short set of smart energy vectors. The model is tested using similarity searches within the Directory of Useful Decoys (DUD) database. The results obtained are considerably better than previously published models. As a Target prediction methodology we propose the use of a Bayesian Classifier that uses a combination of different active compounds to build an energy-dependent probability distribution function for each target.

Figure 1. Estimation of the energy descriptors associated to the methanol.

The original molecule (left) is reduced as a set of discrete atom points (middle). A total of fifteen pairing energies can be estimated from the resulting distribution.

Figure 2. Two-dimensional pairing-energy map for eight DUD targets.

The energy values shown are the most positive and negative pairing energies. As can be appreciated, different clusters are associated to each specific targets.

Figure 3. Comparison of EF at 1% between the USR (green line), ElectroShape (red line), MolShacs (blue line) and the proposed model PED (purple line) with m = 12 and MMFF94 partial charges for ElectroShape and PED.

Figure 4. AUC and EF at 1% for the proposed model (red for m = 12, green for m = 6 and purple for m = 2) and USR (blue line) models.

Averages for different fractions of the training set are estimated (10%, 30%, 50% and 70% of the actives). The origin represents the averaged performance of single similarity retrieving showed at Table 2.

Figure 5. Logarithmic plot of the probability histogram for TPR (blue) and FPR (red) and the difference between them TPR-FPR (light blue).

The Positive Predictive Value (PPV, i.e. precision, in green), and the False Discovery Rate (FDR, purple) are shown in the same graph.