CODES/Neural network model: a useful tool for in silico prediction of oral absorption and blood-brain barrier permeability of structurally diverse drugs

Two different neural network models able to predict both oral absorption (OA) and blood-brain barrier (BBB) permeability of structurally diverse drugs in use clinically are presented here. Using the descriptors generated by CODES, a program which codifies molecules from a topological point of view, we avoid the uncertain choice of molecular conformation and physicochemical parameters. In this work, a method called Reduction of Dimensions, designed for compressing data, is applied for the first time in order to minimize the bias factor added to a QSAR study when the selection of descriptors are performed. A training set of 28 and 35 structurally diverse compounds are used for oral absorption and blood-brain barrier models respectively. The output data is quantitative in both cases and refers to percent of drug absorbed after oral administration (% Bioavailable values) for OA model and log (C-brain/C-blood) (log BB) for BBB permeability model. The network training was completed and validated by the leave-one-out method (Prediction errors were 6.5% and 5.6% for OA and BBB permeability models respectively). Excellent correlations were obtained (r=0.95, r=0.94). Both models show good predictive abilities regarding to external validation on test sets.