Bridging the Gap between Target-Based and Cell-Based Drug Discovery with a Graph Generative Multitask Model.

Area of Science:

• Computational chemistry
• Drug discovery
• Artificial intelligence in medicine

Background:

• Traditional drug discovery relies on target-based screening, which often fails in vivo due to poor drug activity.
• A gap exists between in vitro efficacy and in vivo performance of drug candidates.
• Accurate computational methods are needed to bridge this gap in drug development.

Purpose of the Study:

• To develop a novel graph multi-task deep learning model named MATIC.
• To identify compounds with both target inhibitory (in vitro) and cell active (in vivo) properties.
• To improve the prediction of effective drug candidates for in vivo applications.

Main Methods:

• Developed a multi-task deep learning model (MATIC) using graph neural networks.
• Curated a SARS-CoV-2 dataset for training and validation.
• Investigated model interpretability using molecular property correlations and atom functional attention.
• Employed a Monte Carlo-based reinforcement learning model for generative tasks.

Main Results:

• The MATIC model demonstrated superior performance in screening effective in vivo compounds compared to traditional methods on a SARS-CoV-2 dataset.
• Analysis revealed distinct learned features for in vitro and in vivo tasks.
• Generated novel multiproperty compounds with predicted in vitro and in vivo efficacy.

Conclusions:

• The MATIC model effectively bridges the gap between target-based and cell-based drug discovery.
• The model's interpretability provides insights into drug properties for both in vitro and in vivo activity.
• This approach facilitates the generation of more effective drug candidates with improved in vivo performance.