Correlation of blood-brain penetration and human serum albumin binding with theoretical descriptors

Quantitative Structure-Activity Relationship (QSAR) models were developed for blood-brain barrier and human serum albumin binding for a dataset of drugs where experimental values of both properties were available. All drugs were represented by chemical descriptors calculated from their constitutional, geometrical and topological structure, and quantum mechanical wave function. The obtained linear (multilinear regression) and nonlinear (artificial neural network) models link the drug structures to their reported properties. Also, based on the characterization of the descriptors we suggest additional criteria for the search of new active compounds. Each multilinear model was tested by leave-one-out and ABC methods. The latter method separates the all data points into three sets and predicts for each of them the property values. The former method is an iterative procedure which in each step it excludes one data point and predict its value based on the model rebuilt for the remaining data points. In addition, the predictive ability neural networks were assessed using the validation sets. All drug structures were investigated by conformational analysis in order to find the lowest energy conformers.