Outlier Detection in QSAR Modeling of the Biological Activity of Chemicals by Analyzing the Structure–Activity–Similarity Maps

Abstract

A new method for the detection of outliers in training sets used in QSAR model construction is developed. The method is based on the analysis of structure–activity–similarity (SAS) maps. It involves an empirical assessment of the likelihood of a chemical compound appearing in a particular SAS area. We propose to regard the compounds that have the maximal probability of an “activity cliff” (AC) region and the minimal probability of appearing in a “smooth region” (SR) as outliers. The method proposed can be used in the field of medicinal chemistry to search for new promising biologically active chemical compounds.