Computational modelling of the antischistosomal activity for neolignan derivatives based on the MIA-SAR approach

Abstract

Theoretical models for exploring the antischistosomal activity of a dataset of 18 synthetic neolignans are built using the multivariate image analysis applied to structure–activity relationships (MIA-SAR) approach. The obtained models were validated using the accuracy (Acc) in leave-one-out cross-validation, external validation and Y-randomization procedures, yielding correct classification superior to 80%, 70% and 60%, respectively. Additionally, a comparison was made of the models obtained from binary (black and white) and coloured images; the colours (pixel values) were selected to correspond to chemical properties. It was observed that the models obtained from coloured images with pixel values corresponding to electronegativity (known as the aug-MIA-SARcolour approach) generally yielded superior statistical parameters compared with those obtained from binary images (MIA-SAR) and randomly coloured images (atoms are coloured according to their type) with atomic sizes corresponding to Van der Waals radius (aug-MIA-SAR), respectively. Mechanistic interpretation of the influence of different substituents on the antischistosomal activity revealed that methoxy substituents in the R1 (or R2) and R5 positions of the neolignan scaffold are indispensable for the antischistosomal activity. The obtained results provide knowledge of the possible structural modifications to yield novel neolignan compounds with antischistosomal activity.