TopMT-GAN: a 3D topology-driven generative model for efficient and diverse structure-based ligand design

Area of Science:

• Computational chemistry
• Drug discovery
• Artificial intelligence in medicine

Background:

• 3D structure-based molecular generative models accelerate drug design.
• Generating large, diverse, and effective molecular libraries is challenging.
• Limited comparisons exist between generative models and high-throughput virtual screening (HTVS).

Purpose of the Study:

• Introduce Topology Molecular Type assignment (TopMT-GAN), a novel generative approach for direct structure-based drug design.
• Enable efficient generation of diverse and potent ligands with precise 3D poses.
• Evaluate TopMT-GAN's performance against HTVS methods.

Main Methods:

• Utilized Generative Adversarial Networks (GANs) in a two-step process.
• Step 1: Constructed 3D molecular topologies within a protein pocket using one GAN.
• Step 2: Assigned atom and bond types using a second GAN.

Main Results:

• TopMT-GAN demonstrated robust performance across five diverse protein pockets.
• Achieved a potential enrichment of up to 46,000-fold compared to traditional HTVS.
• Successfully generated diverse and potent ligands with accurate 3D poses.

Conclusions:

• TopMT-GAN is a powerful tool for structure-based molecular generation.
• The method significantly outperforms traditional HTVS in hit and lead generation.
• TopMT-GAN shows great potential for expediting early-stage drug discovery.