3CLpro inhibitors: DEL-based molecular generation.

Area of Science:

• Drug Discovery and Development
• Computational Chemistry
• Medicinal Chemistry

Background:

• Machine learning (ML)-driven molecular generation accelerates drug optimization for known targets.
• Existing methods struggle with novel targets lacking extensive bioactivity data.
• DNA-encoded libraries (DEL) provide systematic, target-specific data, even for new targets.

Purpose of the Study:

• To overcome limitations in molecular generation for novel targets.
• To leverage DEL data for enhanced molecular generation and structural optimization.
• To investigate the impact of transfer learning on molecular generation models using DEL data.

Main Methods:

• Generated 2.96 million structure-affinity samples for 3C-like protease (3CLpro) using an in-house DEL platform.
• Employed molecular docking and an affinity model trained on DEL data to assess transfer learning.
• Filtered generated molecules based on physicochemical properties, drug-likeness, pharmacophore, and molecular docking.

Main Results:

• Successfully generated molecules for a novel target using DEL data, bypassing public databases.
• Demonstrated that transfer learning significantly enhances the positive rate of molecular generation models.
• Identified promising drug candidates through multi-stage filtering and validated by molecular dynamics simulations.

Conclusions:

• DEL platforms offer a powerful solution for generating molecular data for novel drug targets.
• Transfer learning is a key strategy to improve the efficiency and success rate of ML-based molecular generation.
• This integrated approach accelerates the discovery and optimization of drug candidates for challenging targets.