COMPUTER-GUIDED IDENTIFICATION OF NEW INHIBITORS OF CLOSTRIDIUM DIFFICILE SPORE GERMINATION

CARAM FRANCO; TREJO FERNANDO M; PEREZ, PABLO F.; TALEVI ALAN; BELLERA CAROLINA

Congreso; REUNIÓN ANUAL DE SOCIEDADES DE BIOCIENCIAS; 2022; 2022

Clostridium difficile (CD) is a Gram-positive anaerobic spore-forming bacteria. It is the main pathogen causing antibiotic-associated colitis and 20 to 30% of cases of antibiotic-associated nosocomial diarrhea. Meanwhile the global morbidity and mortality rates have been increasing. In this work we developed machine learning models to be used in virtual screening in order to identify new CD spore germination inhibitors.We have compiled 121 molecules tested as inhibitors of CD spores germination, these were divided into three representative groups: training set and two validation groups, using SOMMOC tool. By a combination of feature bagging approximations and forward stepwise we have inferred 1000 ligand-based classificatory models able to recognize inhibitors. Forward, the best individual classifiers model were combined into meta-classifiers and then evaluated by a retrospective screening campaign, against the two validations groups that contained decoys obtained using the LUDe application. The best ensemble model was applied in a prospective virtual screening campaign of DrugBank 5.1.6 (DB), estimating the Positive Predictive Value (PPV) for each compound screened.The training set was balanced with 30 inhibitor compounds and 30 non inhibitors, the validation sets contain, both of them, the same 13 inhibitors, and 24 non inhibitors. The 16 top models with the best performance were combined using the MIN-SCORE operator to improve the predictive capacity and robustness. By analyzing PPV surfaces, a cutoff value of 0.28 was chosen, associated with a specificity of 0.96 and a PPV value of 0.47 for a hypothetical yield of active compounds of 1%.This ensemble model was applied in a virtual screening campaign of DB, obtaining 62 possible hits. As an additional selection criteria, we selected those approved or investigational, obtaining 29 hits.These drugs will be evaluated in in vitro assays and those that are most promising will be tested in animal models.