In silico modeling for the search of new inhibitors of Poly (ADP-ribose) glycohydrolase (PARG) as potential antichagasic drugs

FRANCO CARAM; SALOMÉ VILCHEZ LARREA; LUCAS NICOLÁS ALBERCA; SILVIA H. FERNÁNDEZ VILLAMIL; ALAN TALEVI

Buenos Aires

Congreso; Drug Discovery for Neglected Diseases International Congress 2018 - 4th Scientific Meeting of ResNet NPND; 2018

ResNet NPND

Poly-(ADP-ribose) polymerases (PARPs) are involved in multiple cellular processes such as DNA repair and replication,transcription regulation and apoptosis [1]. The mammalian host PARG proved to be essential for the successful infectionby Trypanosoma cruzi, as inhibition or silencing of human PARG (huPARG) markedly diminished in vitro cell infections.These observations make them an interesting and novel target for the search of new treatments for Chagas disease [2].Objective: Development of computational models capable of identifying new inhibitors of huPARG, and theirsubsequent application to virtual screening (VS) campaigns.We have compiled a database of molecules tested against huPARG. From such database, using a semicorrelationapproach [3] and a random subspace approximation [4] we have inferred 1000 ligand-based classificatory modelscapable of recognizing huPARG inhibitors. These models were validated in a retrospective virtual screening experiment,by seeding a small number of known inhibitors among a large number of putative inactive compounds (DUDe-Library)[5]. The model performance was then evaluated by determination of different metrics: AUROC, RIE and BEDROC [6].Different ensemble learning schemes were tested using a second retrospective screening step (DUDe-2 Library), inorder to improve the individual model?s predictive ability. The best model ensemble was applied in the prospective VSof DrugBank 5 database, estimating the Positive Predictive Value (PPV) for each hit.26 individual classifiers achieved AUROCs above 0.800 against the DUDe-2 Library; among them, only 5 obtainedBEDROC values above 0.30. The highest RIE was 10. Remarkably, the best model ensemble achieved an impressiveAUROC = 0.979, BEDROC = 0.780 and RIE = 70.02. The ensemble was used for prospective VS, selecting 26 alreadyknown drugs with PPV > 20%.Chagas disease mainly affects low-income people from Latin-American countries; the development of new treatmentshas historically been neglected by pharmaceutical industry and most ongoing drug discovery projects depend onfunding from the public sector and non-for-profit organizations. Therefore, time- and cost-efficient strategies such ascomputer-guided drug discovery and drug repositioning are fundamental to expedite the development of novel, moreefficient medications.We have generated a ligand-based model ensemble capable of recognizing huPARG inhibitors, with excellent behaviorin the in silico validation step. The ensemble was applied in VS campaign, finding 26 drugs that could potentially bestarting points for the development of novel antichagasic drugs.