MTMol-GPT: De novo multi-target molecular generation with transformer-based generative adversarial imitation learning

Area of Science:

• Computational Chemistry
• Drug Discovery
• Artificial Intelligence

Background:

• Deep generative models show promise in de novo drug design but often focus on single targets.
• Complex diseases involve multiple factors, necessitating multi-target drugs for enhanced efficacy and overcoming resistance.
• Existing models struggle to generate molecules that simultaneously address multiple disease targets.

Purpose of the Study:

• To develop an advanced generative model for designing multi-target molecules.
• To improve the efficacy and applicability of drug discovery for complex diseases.
• To address limitations in current deep generative models for drug design.

Main Methods:

• Proposed MTMol-GPT, an upgraded Generative Pre-trained Transformer (GPT) model incorporating generative adversarial imitation learning.
• Employed a dual discriminator model with Inverse Reinforcement Learning (IRL) for concurrent multi-target molecular generation.
• Validated the model's performance through extensive generation, molecular docking, and pharmacophore mapping experiments.

Main Results:

• MTMol-GPT successfully generated valid, novel, and effective multi-target molecules for diverse complex diseases.
• Demonstrated robustness and generalization capabilities in molecular generation.
• Generated molecules showed promising drug-likeness properties for potential neuropsychiatric interventions and were applied in a breast cancer drug design case study.

Conclusions:

• MTMol-GPT offers a powerful and broadly applicable solution for multi-target molecular generation in drug discovery.
• The model provides new insights for enhancing therapeutics for complex diseases through high-quality multi-target molecule design.
• This approach signifies a step forward in addressing intricate disease mechanisms and resistance.