DeLA-DrugSelf: Empowering multi-objective de novo design through SELFIES molecular representation

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Automated de novo molecular design is crucial for accelerating drug discovery.
• Existing methods like DeLA-Drug utilize SMILES notation, which has limitations in molecular representation.
• Advancements are needed for robust and data-driven lead optimization strategies.

Purpose of the Study:

• Introduce DeLA-DrugSelf, an upgraded de novo molecular design tool.
• Enhance automated multi-objective drug design capabilities.
• Improve molecular generation using SELFIES (SELF-referencing Embedded String) representation.

Main Methods:

• Employed SELFIES for molecular representation, enabling substitutions, insertions, and deletions.
• Addressed the SELFIES collapse issue by generating only collapse-free compounds.
• Utilized a Pareto dominance-based fitness function within a genetic algorithm for multi-objective optimization.
• Focused on optimizing properties for cannabinoid receptor 2 (CB2R) ligands.

Main Results:

• DeLA-DrugSelf demonstrated significant improvements in drug-likeness, uniqueness, and novelty compared to its predecessor.
• The algorithm effectively handles data-driven scaffold decoration and lead optimization.
• Generated molecules showed enhanced quality metrics.
• Successful application in optimizing CB2R ligands.

Conclusions:

• DeLA-DrugSelf represents a substantial advancement in automated multi-objective de novo molecular design.
• The use of SELFIES and collapse-free generation improves molecular quality and design efficiency.
• The tool is valuable for data-driven optimization of bioactive molecules and lead compounds.
• DeLA-DrugSelf is accessible via a user-friendly web platform for broader application in drug discovery research.