GENERA: A Combined Genetic/Deep-Learning Algorithm for Multiobjective Target-Oriented De Novo Design

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Angiotensin-converting enzyme 2 (ACE2) is a key target in various pathologies, including COVID-19.
• De novo molecular design aims to create novel compounds with specific properties.
• Existing methods may lack efficiency in generating optimized drug candidates for specific targets.

Purpose of the Study:

• Introduce GENERA, a novel algorithm integrating deep learning (DeLA-Drug) and genetic algorithms (GA).
• Assess GENERA's capability for target-oriented de novo design using the ACE2 target.
• Demonstrate GENERA's effectiveness in multiobjective optimization for drug candidate generation.

Main Methods:

• GENERA combines DeLA-Drug for analogue generation with a GA for property optimization.
• The algorithm was applied to the ACE2 target.
• Docking simulations using PLANTS and GLIDE were employed for assessment.
• A fitness function based on Pareto dominance of docking scores was utilized.

Main Results:

• GENERA successfully generated focused molecular libraries targeting ACE2.
• The algorithm demonstrated effective multiobjective optimization using a Pareto dominance-based fitness function.
• Generated libraries showed improved docking scores compared to known ACE2 binders.
• GENERA rapidly produced promising drug candidates.

Conclusions:

• GENERA represents an innovative approach to target-oriented de novo drug design.
• The integration of a DL-based analogue generator within a GA framework is a novel strategy.
• This method efficiently generates optimized molecular libraries for specific biological targets.