“Advantages of a descriptors selection method with múltiple solutions applied to QSPR models for ADMET properties prediction”

DAMIAN PALOMBA; MONICA F. DIAZ; GUSTAVO E. VAZQUEZ

Cordoba

Congreso; 1ª Reunión Internacional de Ciencias Farmacéuticas (RICIFA 2010); 2010

UNC

QSPR (Quantitative structure-property relationship) methods are widely used for ADMET (Absorption, Distribution, Metabolism, Elimination, Toxicology) properties prediction in drug discovery. However, there are some critical issues, which are difficult to address. The usual problems are the selection of the proper number of descriptors that model a desired property, redundant descriptors, chance correlation, overfitting, and getting complex models with low predictive capacity. A method that avoids these problems and in addition provides several models would allow to interpret them using physicochemical principles, and to choose the most suitable one. This work uses a technique based on evolutionary algorithms (1), which provides different predictive models as output. Validity is analyzed for estimating human intestinal absorption (HIA).