Application of Data-Centric Supervised Machine Learning to Predict Phenotypic Activity Against Clinically Relevant Stages of Trypanosoma cruzi

Abstract

Background/Objectives: Chagas disease is a neglected tropical disease caused by theprotozoan parasite Trypanosoma cruzi, which currently affects around 8 million peopleworldwide. The therapeutic arsenal against T. cruzi is so far limited to only two approveddrugs, benznidazole and nifurtimox, that have considerable side effects and limited efficacyin the chronic stage of the disease. Here, we have resorted to supervised phenotypicmachine learning models to explore drug repurposing opportunities and identify potentialnewtherapeutic solutions for Chagas disease. Methods: More than 100,000 bioactivity datapoints were retrieved from ChEMBL and carefully curated according to the data-centricmachine learning paradigm. After curation, two datasets comprising 344 compounds testedagainst T. cruzi Y strain trypomastigotes and 785 compounds tested against Tulahuen strainamastigotes were obtained and used to infer ensemble learning models with excellent average and early enrichment metrics in retrospective screening experiments (AUROC > 0.96and EF0.01 > 58). A prospective screening campaign was then performed on DrugBankand the Drug Repurposing Hub databases, submitting eight in silico hits for experimentalconfirmation. Results: Six of the in silico hits confirmed their predicted trypanocidal effects.Conclusions: We havebuilt portable meta-classifiers capable of identifying small moleculeswith trypanocidal activity against amastigotes, the clinically most relevant stage of T. cruzi.The predictive ability of this meta-classifier was experimentally validated.