Machine learning-based screening of in-house database to identify BACE-1 inhibitors

The beta-site APP cleaving enzyme-1 (BACE-1) is one of the key targets for novel drugs to treat Alzheimer's disease (AD). The BACE-1 plays a key role in the amyloidogenic process, leading to the production of amyloid-beta (A beta) plaques in the brain. In the present work, we have developed an ML model based on the sulfonamides dataset. The best ML model was built using the XGBoost algorithm on PubChem fingerprints. The model had an accuracy, precision, recall and F1 score of 0.89, 0.88, 0.99 and 0.93, respectively, on the validation set. The same model was used to screen the database of previously synthesized and reported in-house compounds. The screening resulted in the identification of two hits, i.e., compound 28 and compound 37. Both the compounds were screened for their BACE-1 inhibitor activity. The IC50 value of compound 28 was found to be 0.431 +/- 0.006 mu M, and compound 37 showed an IC50 value of 0.272 +/- 0.019 mu M. The docking study revealed that compound 37 also showed interactions with the catalytic dyad of BACE-1, i.e., Asp32 and Asp228.