In silico modeling for the search of new inhibitors of falcipaine 2 of Plasmodium falciparum

TALEVI A.; MORALES J.F.; ALBERCA L.N.

Santiago

Congreso; XXIV Congreso Latinoamericano de Parasitología; 2017

Sociedad Chilena de Parasitología (SOCHIPA)

Malaria is a mortal disease caused by Plasmodium parasites that are transmitted through the bites of infected anopheles mosquitoes. Approximately half the world´s population live in malaria endemic areas. According to the World Health Organization estimates (December 2016) there were 212 million cases of malaria in 2015 and 429,000 deaths. The most virulent human malaria parasite is Plasmodium falciparum. Despite there exist antimalarial drugs used clinically, the control of malaria is difficult because of the widespread resistance of P. falciparum to the available drugs. For this reason, it is necessary to find new treatments. An interesting strategy for the search of new treatments is the computer-assisted drug repurposing, that consists in finding new therapeutic indications for existing drugs through the application of computational models. Falcipains are a group of cysteine proteases present in P. falciparum that participate in several vital processes of the erythrocytic cycle of the malaria. One of these proteins, Falcipain 2 (FP-2), has been reported as an interesting drug target due to the functions carried out during the hemoglobin degradation pathway. Objective: Finding new inhibitors of FP-2 by computer-guided drug repositioning. Methodology: By an extensive literature search we have compiled a database of chemical compounds tested against FP-2. This database was used to generate classificatory models capable of identifying FP-2 inhibitors among already approved drugs. The models were generated by machine learning methods using the R environment and validated through analysis of the Area under the ROC curve. We have generated a more extensive database containing a higher number of inactive compounds (DUDe database) to validate the models in a more real situation. The best models were ensembled by four operators and the best model ensemble was used for the virtual screening of Sweetlead and Drugbank databases. Finally, we have determined the probability that each compound will be active against FP-2 by determining of Positivity Predictive Values (PPV). Results: 31 of the 1000 generated individual models obtained AUROCs between 0.80 and 0.85 against the DUDe database. The model ensembles considerably improved these results. Particularly, the Minimum ensemble of the best 11 models obtained an AUROC of 0.92, thus we used this ensemble for the virtual screening. We have found 11 approved drugs with PPV > 25% in DrugBank database. Conclusion: The use of computational models has allowed identifying potential new FP2 inhibitors among already approved drugs.