MIA-plot: a graphical tool for viewing descriptor contributions in MIA-QSAR

A major challenge faced with the MIA-QSAR (Multivariate Image Analysis applied to Quantitative Structure-Activity Relationships) technique in molecular modeling is to obtain chemically intuitive information from the molecular descriptors that correlate to bioactivity values. This work reports a graphical tool that uses PLS regression coefficients ( b ) and the variable importance in projection (VIP) scores to give insight on the structural motifs responsible for increased or attenuated biological activities in a congeneric series of compounds. Three sets of compounds useful in food chemistry, namely fatty acids and derivatives (antimicrobial), anilides (antimicrobial) and disaccharides (sweeteners), were used to assess the practical contribution of these parameters in yielding more interpretable models. The MIA-QSAR models demonstrated satisfactory prediction performance and, most importantly, the relevance of single MIA predictors to the response variable could be assessed, thus allowing for the structural, electronic and physicochemical interpretation of the MIA-QSAR models. While VIP-based graphs were especially useful in providing insight on the molecular regions/substituents most affecting the property values, the b -based analysis indicated whether the particular regions/substituents negatively or positively influenced the dependent variables. It is therefore recommended that these parameters be employed as complementary tools for a more robust and lucid model interpretation. Chemically intuitive information from MIA molecular descriptors that correlate to bioactivity values was obtained from the analysis of PLS regression coefficients and variable importance in projection scores.