Exploratory Data Analysis of the In Vitro Effects of Novel Hydrazide-Hydrazone Antioxidants in the Context of In Silico Predictors.

Substantial in vitro experimental data have been produced about the safety, antioxidant, neuro- and hepatoprotective effects of a series of recently synthesized N-pyrrolyl hydrazide-hydrazones (compounds , -). However, compound activity across multiple assays varies and it is challenging to elucidate the favorable physicochemical characteristics of the studied compounds and guide further lead optimization. The aim of the current study is to apply exploratory data analysis in order to profile the biological effects of the novel hydrazide-hydrazones, gain insights related to their mechanisms of action in the context of in silico predictions and identify key predictor-outcome relationships. We collected a dataset from available in vitro studies of compounds , -. It included cytotoxicity values, protection against hydrogen peroxide-induced damage in HepG2 and SH-SY5Y cells, two radical scavenging assays and a hemolysis assay across a range of treatment concentrations. SwissADME-based predictions of chemometric and ADME parameters and pro-oxidant enzyme docking data were generated to provide context for the interpretation of in vitro outcome patterns and identify causal relationships. Multiple factor analysis (MFA), followed by hierarchical clustering on principal components (HCPC), was applied to profile compounds' biological behavior. This revealed that differences in the number of H-bond donors, in the permeability coefficient and in the docking scores to two pro-oxidant enzymes could aid in explaining the effects of compounds with similar in vitro profiles. HCPC differentiated as mostly neuroprotective, and as hepatoprotective radical scavengers, with higher docking affinity to 5-lipoxygenase (5-LOX) and myeloperoxidase (MPO) and , and as having less H-bond donors and variable in vitro activity. The consensus application of three variable selection approaches based on standard lasso regression, robust penalized regression and random forest confirmed the relationships between some in vitro outcomes and LogP, pan-assay interference (PAINS) alerts, 5-LOX allosteric site docking and H-bond donor numbers. The exploratory analysis of the combined in vitro and in silico dataset provides useful insights which could help explain the major drivers behind the experimental results. It can be informative in the design of new, improved members of the series of novel N-pyrrolyl hydrazide-hydrazones with better neuroprotective potential and less side effects.