Computer-aided drug repositioning focused on inhibitors of Trypanothione synthetase

ALICE JUAN IGNACIO; MORALES JUAN FRANCISCO ; DIEGO BENÍTEZ; LUCIANA GAVERNET; MARCELO COMINI; ALAN TALEVI; CAROLINA L. BELLERA

Buenos Aires

Congreso; DRUG DISCOVERY FOR NEGLECTED DISEASES INTERNATIONAL CONGRESS 2018; 2018

/ Instituto de la Química y Metabolismo del Fármaco FFyB UBA

Computer-aided drug repositioning may contribute to the systematic identification of second medical uses for existing drugs, allowing the efficient discovery of potential therapeutic solutions for neglected conditions. Trypanothione synthetase (TryS) is an essential enzyme for the maintenance of the redox balance in trypanosomatids [1,2].Objective: Building ligand-based in silico models capable of identifying novel TryS inhibitors, for subsequent application in virtual screening (VS)experiments.A dataset of 401 compounds assayed against TryS were compiled, including an in house library (lab. Redox Biology of Trypanosomes, Institut Pasteur de Montevideo). It was used to generate: a balanced training set; an independent test set and; two libraries (DUDA and DUDB) later used for retrospective VS campaigns, in which a small number of known inhibitors were seeded among a large number of decoys extracted from the enhanced Directory of Useful Decoys[3]. Using a semicorrelation approach [4] random subspace approximation [5], 1000 ligand-based classificatory models were built. Ensemble learning was later used to obtain a 90-model combination which was finally applied in the VS ofDrugBank and Sweetlead databases.The best individual model obtained a remarkable area under the receiver operating characteristic curve (AUROC) of 0.982 for the training set, 0.946 for DUDe-1 and 0.964 for DUD-e2 (AUROC = 1 denotes a perfect classifier). The90-model ensemble achieved an AUROC of 0.977 for DUDA and 0.938 for DUDB, and more consistent behavior in the correspondent Positive Predictive Value (PPV) surfaces. Analysis of PPV surfaces led us to choose a score cutoff valueassociated to a sensitivity/specificity ratio of 0.88 and a PPV of 24% for a hypothetic active yield of 1%. 25 hits were selected using that cutoff during the prospective (real) VS.Today, funding for drug discovery projects focused on neglected conditions still depends, mostly, on the public sector and non-for-profit organizations. Computer-aided drug discovery and computer-guided drug repurposing representcost- and time-efficient approximations to accelerate the development of innovative medications for such conditions. We have generated a ligand-based 90-model ensemble capable of recognizing TryS inhibitors, with outstandingbehavior in the in silico validation step. Here, ensemble learning was shown to improve the robustness and generalization ability of computational models. The best ensemble was applied in VS campaign, finding 25 hits that uponconfirmation of their inhibitory activity against the enzyme and parasites may be potentially used as starting points to develop new target-based antichagasic treatments.