De novo design of anti-tuberculosis agents using a structure-based deep learning method

Area of Science:

• Drug discovery and development
• Computational chemistry
• Infectious diseases

Background:

• Mycobacterium tuberculosis (Mtb) poses a significant global health threat due to widespread antibiotic resistance.
• Existing anti-TB treatments are becoming less effective, necessitating the development of novel therapeutic agents.
• Targeting essential Mtb-specific proteins is a key strategy for overcoming drug resistance.

Purpose of the Study:

• To design novel anti-tuberculosis agents targeting the Mtb chorismate mutase protein.
• To develop and apply a structure-based deep learning approach for small molecule generation.
• To overcome the limitations of traditional drug design methods that require extensive target-specific ligand data.

Main Methods:

• Utilized a structure-based deep learning algorithm for conditional small molecule generation.
• Employed a graph attention model to identify key binding site residues of Mtb chorismate mutase.
• Focused solely on the protein's binding site structure, eliminating the need for large ligand datasets.

Main Results:

• Generated novel small molecules with high complementarity to the Mtb chorismate mutase binding site.
• Identified key residues influencing molecular design through the graph attention model.
• Proposed molecules with pharmacophoric features similar to known inhibitors, suggesting potential bioactivity.

Conclusions:

• A structure-based deep learning method can effectively design novel drug candidates for challenging targets like Mtb.
• This approach accelerates drug discovery by reducing reliance on extensive pre-existing experimental data.
• The designed molecules represent promising leads for developing new anti-tuberculosis therapies against drug-resistant strains.