Development of machine learning solutions for targeted exploitation of enormous chemical space and objective-guided molecular design are paramount in modern biomedical research and have gained a significant momentum. We present mathematical formalisms and implementation of generative machine learning strategies for objective-guided molecular design and cross-domain transformation of small molecules to the domain of selective probes of protein kinases. We developed a portfolio of machine learning approaches that include Generative Adversarial Networks (GAN) for objective-guided molecular design of specific chemical probes targeting the ABL and SRC kinase; (b) de novo generative machine learning approach with Bayesian optimization and chemical perturbation; (c) generative perturbation learning for drug repurposing into targeted specific kinase probes. For generation of a comprehensive dataset of kinase chemical probes used in machine learning training, we consolidate prioritize and integrate known kinase inhibitors from 10 major kinase families along with small molecule databases including ZINC and the GDB-17. After encoding and mapping molecules in a continuous latent space representation, we develop a robust chemical feature-based machine learning predictor of kinase inhibition likelihood that is used in perturbation-based transformation of small molecules. The developed perturbation-based generative design approaches allow for an efficient manipulation and transformation of chemical space into predefined classes of specific kinase probes. By combining molecular perturbation design with the kinase inhibition likelihood analysis and similarity assessments, we demonstrate how autonomous molecular design strategy can morph inhibitors of LCK, ABL1, and EGFR kinases into selective chemical probes of SRC kinase. We will discuss how integration of experiment-informed protein modeling, network science and generative machine learning approaches could lead to a new general strategy for autonomous molecular design and validation of targeted and allosteric chemical probes with applications to families of protein kinases and molecular chaperones.

Autonomous molecular design and targeted transformations of small molecule space into specific protein kinase inhibitors using hierarchical perturbation-based machine learning approaches

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

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