Ensemble learning application to discover new trypanothione synthetase inhibitors

Abstract

Trypanosomatid-caused diseases are among the neglectedinfectious diseases with the highest disease burden, affecting about 27 millionpeople worldwide and, in particular, socio-economically vulnerable populations.Trypanothione synthetase (TryS) is considered one of the most attractive drugtargets within the thiol-polyamine metabolism of typanosomatids, being unique,essential and druggable. Here, we have compiled a dataset of 401 T. brucei TrySinhibitors that includes compounds with inhibitory data reported in theliterature, but also in-house acquired data. QSAR classifiers were derived andvalidated from such dataset, using publicly available and open-source software,thus assuring the portability of the obtained models. The performance androbustness of the resulting models were substantially improved through ensemblelearning. The performance of the individual models and the model ensembles wasfurther assessed through retrospective virtual screening campaigns. At last, asan application example, the chosen model-ensemble has been applied in aprospective virtual screening campaign on DrugBank 5.1.6 compound library. Allthe in-house scripts used in this study are available on request, whereas thedataset has been included as supplementary material.