Combining machine learning models and 3D conformational structure to prioritize novel leads against Trypanosoma cruzi from a chromane scaffold library

MERCEDES DIDIER GARNHAM; LUZ SOMMARIVA; EMIR SALAS SARDUY; FERNAN AGUERO

La Plata

Congreso; XXXIV Reunión Anual de la SAP 2023; 2023

Sociedad Argentina de Protozología

We have recently developed a pipeline to map conserved druggable modules between yeasts and T. cruzi, starting from a number of genome‐wide fitness profiling assays performed on S. cerevisiae. From these experiments we obtained information on the cellular response of yeast mutant strains to different chemicals. Because each mutant carries a single and traceable heterozygous deletion, this strategy provides a rich source of pharmacogenomic associations between drugs and genes, which are good starting points to guide repurposing opportunities. Using a bioinformatic pipeline, we analyzed 271,995 gene-drug chemogenomic pairings and identified 196 suitable candidates for repurposing to T. cruzi, from which 21 compounds were tested in phenotypic screenings and 3 showed micromolar potencies. One of the active compounds (CID_18) was a chromane analog, a scaffold reported as a promising lead for anti-protozoal drugs (Jadhav et al, 2023). Starting from this compound, we generated a library of 17,946 bioactive compounds with a chromane scaffold. We prioritized candidates using two approaches. First, we analyzed the 3D conformational structure of the active compound using vROCS and selected 500 compounds from the chromane library with the most similar structure. Second, we trained 3 machine learning classifier models using 15 physicochemical properties of compounds with reported activity (or inactivity) on T. cruzi and human targets to predict activity against T. cruzi and discard undesired interactions in humans. The best model for each organism was applied to the complete chromane library to predict their respective activity. As a result, we obtained 131 lead compounds with a chromane scaffold, shape and distribution of chemical features similar to CID_18, and with predicted activity and selectivity on T. cruzi. Next, we will curate and acquire candidates from this list which will be assayed in vitro, against T. cruzi amastigotes.