Beta-secretase 1 (BACE1) is an enzyme involved in production of amyloid-β peptides and BACE1 inhibitors have been evaluated in clinical trials for Alzheimer’s disease. The AIDD module in ADMET Predictor<sup>®</sup> is a compound design technology that generates candidate compounds, scores them using various objective functions, and selects Pareto optimal molecules for the next iteration. Candidate compounds are generated from a seed molecule(s) via transformations (encoded as SMIRKS strings) that are medicinal chemistry aware, i.e., unstable, toxic, and non-druglike compounds are avoided. Penalties are applied to predictions that are outside the models’ applicability domain to de-prioritize these compounds. The algorithm is extremely fast, more than 5 million candidates can be evaluated in 24 hours on an Intel® Core™ i7 CPU (1.8 GHz), 4-core processor using 8 threads.<br/><br/>Objective functions for design of BACE1 inhibitors included a BACE1 QSAR model, ADMET liability, synthetic feasibility, blood brain barrier penetration, and oral bioavailability. The BACE1 QSAR model is an artificial neural network ensemble model trained on a data set of 370 BACE1 inhibitors whose pIC50 values ranged from 3.7 to 9.7. One run of 50 iterations that generated 500 candidate compounds per iteration started with a dihydro-1,4-oxazine containing compound with an IC50 of 6.9 nM. It generated 450 candidate compounds and the most potent compound had a predicted BACE1 IC50 value of 0.11 nM. One of the candidate compounds contained a sulfone group in place of the oxygen in the starting oxazine ring and was an exact match of one of the compounds in the BACE1 data set. Several of the candidate compounds contain a dihydropyrazinone ring in place of the starting structure. Again, one of these compounds is an exact match of a compound in the BACE1 data set. The run generated 43 molecules with predicted BACE1 IC50 values below 10 nM, very low ADMET liability score, good synthetic feasibility, and high predicted oral bioavailability.

Using artificial intelligence to design BACE1 inhibitors

ACS Spring 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.