Evaluating the outputs of ARAE-based molecular generative model for synthetic accessibility and BTK inhibition.
TL;DR
This proof-of-concept study evaluated hypothetical drug-like molecules generated by an adversarial regularized autoencoder (ARAE) architecture for their synthetic tractability and inhibitory effects on Bruton's tyrosine kinase (BTK).
Abstract
The application of artificial intelligence (AI) methodologies in drug discovery represents a rapidly expanding area of interest, providing a valuable means for exploring uncharted chemical spaces to identify novel therapeutic candidates. In this proof-of-concept study, we evaluated hypothetical drug-like molecules generated by an adversarial regularized autoencoder (ARAE) architecture, which were not documented in existing chemical literature and databases, for their synthetic tractability and inhibitory effects on Bruton's tyrosine kinase (BTK). Utilizing a structure-based computational approach, we identified a predicted candidate for BTK inhibition. The lab-based preparation of this AI-generated compound confirmed the synthetic feasibility of the ARAE outputs, although it demonstrated only weak BTK inhibition experimentally. Subsequent structural optimization revealed an analog of the original structure, containing an aniline moiety in place of the piperidine amide, that exhibited modestly improved inhibitory activity. These findings indicate that further development of our ARAE model holds promise for the discovery of novel, synthetically accessible scaffolds relevant to drug discovery.