Synthesizability-constrained expansion and multi-objective evolution of antitubercular compounds

点击翻译按钮获取中文摘要

PROJECT SUMMARY New approaches are urgently needed to advance new chemical entities as antitubercular agents of clinical relevance. A key hurdle is the multiple criteria process that constitutes hit-to-lead optimization of small molecules with demonstrated in vitro potency versus drug-sensitive and drug-resistant strains of the causative bacterium Mycobacterium tuberculosis (Mtb). This project will address the design, construction, and validation of novel machine learning approaches to predict two of these critical molecular properties: in vitro Mtb growth inhibition and mouse pharmacokinetic exposure in the plasma. The resulting models, validated through external test statistics, will be complemented by computational approaches, relying on expert-encoded reaction templates and/or learned reaction prediction models, to predict optimal synthetic routes to two promising antitubercular small molecules: JSF-3005 and CD117. These data will inform the enumeration of synthesizable analogs for hit expansion to be followed by the selection of candidate analogs scored with a multi-objective Pareto optimization criterion combining multiple surrogate QSAR models with or without docking scores. Optimality scores will guide a genetic algorithm for synthesizability-constrained molecular optimization as a replacement for exhaustive forward enumeration. The top-scoring candidate lead compounds in each series will be then assayed for critical molecular properties to validate this novel approach and supply novel antitubercular agents for further study outside the scope of this proposal.