AI structure prediction speeds discovery of 'molecular glues' to treat disease

A Baylor College of Medicine-led research team has developed an approach that merges the examination of thousands of proteins with artificial intelligence to expedite the discovery of small molecules known as molecular glues, which hold promise for treating diseases. The team's findings, published in Nature Communications, reveal a novel class of molecular glues that could potentially neutralize proteins associated with blood cancers and autoimmune disorders.

Dr. Jin Wang, senior author and director of the Center for NextGen Therapeutics, explained that molecular glues function as cellular matchmakers, bringing target proteins to the cell's natural protein-disposal machinery, which subsequently eliminates them. In this study, the researchers focused on the protein VAV1, which plays a crucial role in immune cell function and has been implicated in blood cancers and autoimmune diseases.

By screening a library of molecules using high-throughput proteomics, the team identified several compounds, including NGT-201-12, that effectively lowered VAV1 levels while minimally affecting other proteins. Follow-up tests confirmed that these compounds operate via the cell's natural protein-recycling system, specifically relying on the protein cereblon (CRBN), a vital component of the protein-degradation pathway.

To better understand the molecular glue's interaction with its target, the researchers employed a computational platform called GluePlex, which used artificial intelligence, protein-structure prediction tools, and physics-based modeling to simulate the formation of a three-part complex between VAV1, CRBN, and the molecular glue.

This model identified a specific region within VAV1, termed the SH3-2 domain, as being essential for degradation. Experimental validation confirmed this prediction, pinpointing the exact location on VAV1 where the molecular glue binds—a small surface loop acting as a degradation signal, or "degron." This discovery is significant as it deviates from the common structural feature, known as the G-loop, found in many cereblon-dependent molecular glues.

The finding suggests that cereblon might recognize a broader range of structures, potentially expanding the scope of proteins that can be targeted for degradation. Upon identifying the molecular mechanism, the researchers enhanced the compounds through medicinal chemistry, introducing chlorine atoms to reduce molecular flexibility and increase degradation efficiency.

One of the optimized compounds, NGT-201-18, demonstrated enhanced potency and formed a stronger protein complex necessary for degradation. The team also tested NGT-201-18 in primary human T cells, observing that the compound reduced VAV1 levels and suppressed T-cell activation, indicating that the degrader could modulate immune-cell signaling in a clinically relevant context.

While further research is needed before these compounds can be considered for patient use, the study underscores the potential of VAV1 degradation as a therapeutic strategy for autoimmune and inflammatory diseases. Importantly, the research also emphasizes the value of comprehensive proteomic profiling, as the study discovered that some compounds could degrade an additional protein, LIMD1.

This finding underscores the importance of evaluating both intended and unintended protein targets during drug development. Overall, this work introduces a range of VAV1-targeting molecular glues and demonstrates how AI, structural modeling, and proteomics can accelerate drug discovery at the earliest stages of research, potentially leading to new treatments for immune-related diseases and cancers with limited therapeutic options.