Active Learning to Select the Most Suitable Reagents and One-Step Organic Chemistry Reactions for Prioritizing...

Area of Science:

• Computational chemistry
• Drug discovery
• Medicinal chemistry

Background:

• Identifying novel drug candidates from large chemical spaces is crucial for modern drug discovery.
• Early identification of hits amenable to optimization accelerates the drug development pipeline.

Purpose of the Study:

• To develop a fully automated active learning protocol for proposing chemical reagents and reactions.
• To enumerate target-specific primary hits from ultralarge chemical spaces based on protein binding site structure.

Main Methods:

• Utilizing a protein's 3D binding site structure as the starting point.
• Employing a fully automated active learning protocol to suggest commercial reagents and one-step organic reactions.
• Applying the method in single and multiple transform scenarios to chemical spaces ranging from 670 million to 4.5 billion compounds.

Main Results:

• The protocol successfully recovered up to 98% of virtual hits identified by exhaustive docking.
• The method scanned only 5% of the full chemical space, demonstrating high efficiency.
• Achieved high throughput screening of trillion-sized chemical spaces with minimal computational cost.

Conclusions:

• The developed protocol offers an efficient and computationally inexpensive approach for structure-based screening.
• It enables rapid enumeration of drug-like molecules from massive chemical libraries.
• This method significantly advances the early stages of drug discovery by identifying readily optimizable hits.