First report on predictive chemometric modeling, 3D-toxicophore mapping and in silico screening of in vitro basal cytotoxicity of diverse organic chemicals

Classification and regression based quantitative structure-toxicity relationship (QSTR) as well as toxicophore models were developed for the first time on basal cytotoxicity data (in vitro 3T3 neutral red uptake data) of a diverse series of chemicals (including drugs and environmental pollutants) collected from the ACuteTox database (http://www.acutetox.eu/). Statistically significant QSTR models were obtained using linear discriminant analysis (classification) and partial least squares (regression) methodologies. Generated toxicophore models showed four important features responsible for basal cytotoxicity: (i) two hydrophobic aliphatic groups (HYD Aliphatic), (ii) ring aromatic group (RA) and (iii) hydrogen bond donor (HBD). The most predictive hypothesis (Hypo 1) had a correlation coefficient of 0.932 for the training set, a low rms deviation of 1.105, and an acceptable cost difference of 62.8 bits, which represents a true correlation and a good predictivity. QSTR and toxicophore models were rigorously validated internally as well as externally along with the randomization test to nullify the possibilities of chance correlation. Our in silico models enable to identify the essential structural attributes and quantify the prime molecular pre-requisites which were chiefly responsible for in vitro basal cytotoxicity. The developed models were also implemented to screen basal cytotoxicity for huge number DrugBank database (http://www.drugbank.ca/) compounds.

► Basal cytotoxicity is a must-have data for the approval of a new drug. ► QSTR and toxicophore models were developed for the ACuteTox database chemicals. ► Features identified by QSTR models were in agreement with toxicophore and PDD. ► This approach offers an efficient screening tool for the in vitro basal cytotoxicity. ► Developed models were applied to screen basal cytotoxicity for DrugBank database.