Expediting hit-to-lead progression in drug discovery through reaction prediction and multi-dimensional optimization

Area of Science:

• Medicinal Chemistry
• Drug Discovery
• Computational Chemistry

Background:

• Accelerating the synthesis of novel bioactive compounds is crucial for drug discovery.
• The hit-to-lead optimization phase is a critical bottleneck in developing new therapeutics.

Purpose of the Study:

• To demonstrate an integrated workflow combining high-throughput experimentation (HTE) and deep learning to accelerate hit-to-lead optimization.
• To identify novel, potent inhibitors of monoacylglycerol lipase (MAGL) through computational design and experimental validation.

Main Methods:

• Generated a dataset of 13,490 novel Minisci-type C-H alkylation reactions using HTE.
• Trained deep graph neural networks to predict reaction outcomes.
• Performed scaffold-based enumeration to create a virtual library of 26,375 molecules.
• Evaluated virtual library using reaction prediction, physicochemical properties, and structure-based scoring.
• Synthesized and characterized 14 lead MAGL inhibitor candidates.

Main Results:

• Identified 212 MAGL inhibitor candidates from the virtual library.
• Synthesized 14 compounds exhibiting subnanomolar activity, a 4500-fold potency increase.
• Achieved favorable pharmacological profiles for the identified inhibitors.
• Obtained co-crystal structures of three designed ligands with MAGL, revealing binding modes.

Conclusions:

• The integrated workflow significantly reduces cycle times in hit-to-lead progression.
• Combining miniaturized HTE, deep learning, and molecular property optimization is effective for drug discovery.
• This approach enables rapid diversification of hit and lead structures for accelerated therapeutic development.