Prediction of the Brain−Blood Distribution of a Large Set of Drugs from Structurally Derived Descriptors Using Partial Least-Squares (PLS) Modeling

Abstract

In this study, the multivariate partial least-squares projections to latent structures (PLS) technique was used for modeling the brain-blood concentration ratio (BB) of 61 structurally diverse compounds. The PLS model was based on molecular descriptors that can be calculated for any compound simply from a knowledge of its molecular structure, and the model included several topological and constitutional descriptors. The PLS analysis resulted in a significant three-component model with the following statistics: r = 0.922, Q = 0.867, s = 0.318, n = 58, and F = 102. The predictive ability of the model was assessed by means of crossvalidation and also by using BB partitioning data, BBB permeability data, and 1 set of qualitative brain penetration data, resulting in BB distribution data for 97 compounds. The results indicate that the PLS model developed is statistically sound and is sufficiently robust for predictive use. Taking into account the great ease of computation and interpretation of the derived model, it may be of general utility in predicting BB ratios for a very wide range of new drugs.