STELLA provides a drug design framework enabling extensive fragment-level chemical space exploration and balanced...

Area of Science:

• Computational Chemistry
• Medicinal Chemistry
• Drug Discovery

Background:

• Traditional drug discovery methods face challenges in exploring vast chemical spaces and optimizing multiple molecular properties.
• Existing generative models often exhibit limitations in efficiency and the breadth of chemical diversity they can explore.

Purpose of the Study:

• To introduce STELLA, a metaheuristics-based generative molecular design framework.
• To demonstrate STELLA's superior performance in chemical space exploration and multi-parameter optimization for drug discovery.

Main Methods:

• STELLA integrates an evolutionary algorithm for fragment-based chemical space exploration.
• It employs a clustering-based conformational space annealing method for multi-parameter optimization.
• Deep learning models are utilized for accurate prediction of pharmacological properties.

Main Results:

• STELLA generated 217% more hit candidates and 161% more unique scaffolds compared to REINVENT 4 in a case study.
• STELLA achieved more advanced Pareto fronts, indicating superior multi-objective optimization.
• In simultaneous optimization of 16 properties, STELLA consistently outperformed control methods in average objective scores and chemical space coverage.

Conclusions:

• STELLA demonstrates superior efficiency in exploring chemical space and achieving multi-parameter optimization.
• The framework offers a powerful tool for de novo molecular design, accelerating the drug discovery process.
• STELLA's capabilities suggest significant advancements in generating optimized drug candidates.