SMILES-based optimal descriptors: QSAR modeling of estrogen receptor binding affinity by correlation balance

Abstract

Quantitative structure—activity relationships for model of estrogen receptor relative binding affinity (pRBA) have been built. These models are one-variable correlations between pRBA and optimal descriptor calculated with simplified molecular input line entry system. These models were obtained by means of the correlation balance: one subset of the training set (sub-training set) plays role of the training; the second subset (calibration set) plays role of the preliminary check of the models. Three splits into the sub-training set, calibration set, and external test set were examined. It has been shown that the correlation balance is a more robust predictor for the endpoint than classic scheme (training set–test set: without of the calibration). The statistical characteristics of the model are n = 59, r2 = 0.8792, s = 0.643, F = 415 (sub-training set); n = 39, r2 = 0.8805, s = 0.637, F = 273 (calibration set); and n = 31, r2 = 0.8132, s = 0.781, F = 126 (test set).