Allosteric protein regulation, achieved through chemical probes/modulators designed to bind to allosteric sites (usually distant from traditional orthosteric sites), enables effective control over protein functional activity. The potential structural diversity of allosteric sites allows for the design of chemical probes with high specificity and selectivity, minimizing potential future toxicity. Therefore, the identification of potentially allosteric binding sites and accessing their druggability is a promising tactic in modern drug discovery. We are developing and applying molecular dynamics (MD)-based, graph-theoretical, and AI-based approaches to study protein allostery in the context of drug discovery.

The rapid expansion of purchasable, on-demand chemical libraries containing billions or even trillions of molecules has greatly advanced modern drug discovery. However, it has also posed significant challenges to the efficient application of traditional structure-based virtual screening methods. To address this challenge, we have been developing a novel computational methodology that greatly accelerates virtual screening. Our workflow uniquely integrates machine learning, generative chemistry, massive chemical similarity searching methods, and minimal molecular docking to nominate a small number of top-scoring virtual hits selected from ultra-large chemical libraries (e.g., 36B Enamine REAL Space). We have observed that our methodology yields the highest enrichment factors over random compound selection compared to state-of-the-art accelerated virtual screening methods, while requiring the least computational resources.

DEL technologies have become increasingly popular as an effective hit and lead generation strategy, substantially reducing the cost per compound tested compared to traditional high-throughput screening. However, the method has several major limitations, including reliance on basic computational and visualization tools for manual data processing and analysis, off-DNA hit resynthesis for further confirmatory screenings and hit prioritization, and efficient combinatorial library design. These constraints can significantly delay the discovery of viable hits and leads. Thus, we are developing and implementing a comprehensive computational approach, in collaboration with Dr. Pearce’s lab, that utilizes DEL screening data and state-of-the-art AI modeling. This approach aims to identify and prioritize diverse, high-quality hits from commercially available compound libraries, as well as de novo generated chemical matter. Additionally, this methodology will help lay the foundation for new strategies to efficiently design target-specific DEL libraries, where AI models can suggest DEL fragments, purchasable or de novo, amenable to certain chemical reactions and more likely to have affinity toward specific families/classes of proteins.

By utilizing both our in-house developed computational tools and publicly available resources, we make significant contributions to collaborative translational projects within the CICBDD. As the center we are dedicated to expanding our knowledge of biological targets and exploring the broader biological consequences that arise from modulating these targets.