APPLICATION OF IN SILICO DRUG REPURPOSING APPROACHES IN THE SEARCH OF NEW ANTICONVULSANT DRUGS ACTIVE IN THE PTZ KINDLING MODEL

ESTEFANÍA PERALTA; MAXIMILIANO FALLICO; DENIS N. PRADA GORI; LUCAS N. ALBERCA; CAROLINA L. BELLERA

Mar del Plata

Encuentro; Reunion Anual de Sociedades de Biociencias; 2022

Sociedad Argentina de Investigación Clínica

Drug repurposing involves search of new medical uses for alreadyknown drugs, including approved, discontinued, shelved and experimentaldrugs. In this work, we have used machine learning approximationsto develop in silico models capable of identifying novel anticonvulsantdrugs with protective effects in the PTZ kindling model.For the generation of algorithms, 162 compounds with and withoutprotective effects in the PTZ kindling model in mice were compiledfrom literature; this dataset was divided into representative trainingand test sets by clustering technique. Afterwards, linear classifiermodels were generated in Python. The best classifiers obtainedwere combined into meta classifiers and validated by retrospectiveselection experiments. As a result, the dataset was partitioned intoa training set of 41 active compounds, 41 inactive compounds, andthe test set of 40 active and 40 inactive compounds. In turn, thistest set was subdivided into two validation groups that were complementedwith 1000 decoys, in order to evaluate the performanceof the models obtained. According to this performance, 3000 linearclassification models were generated, from which the 7 with the bestperformance were chosen and combined using the PROM-SCOREoperator to improve the predictive capacity and robustness of theobtained models. Using positive predictive surface analysis (PPV),the cutoff value of 0.961 associated with a specificity of 0.979 anda PPV value of 0.20 was chosen for a hypothetical yield of activecompounds of 1%. The best ensemble model was applied in a virtualscreening of Drug Bank, Sweet Lead and DRH databases. 90approved drugs were identified as potential protective agents in thePTZ kindling model. The present study constitutes an example ofthe use of machine learning approximations to systematically guidedrug repurposing projects