Structure-Based Generation of 3D Small-Molecule Drugs: Are We There Yet?

Area of Science:

• Computational chemistry
• Drug discovery
• Medicinal chemistry

Background:

• Structure-based drug design (SBDD) is vital for preclinical drug discovery.
• Generative algorithms streamline SBDD by designing molecules based on protein binding pockets.
• A lack of metrics hinders evaluation of generated molecules' chemical plausibility.

Purpose of the Study:

• Introduce novel metrics to assess chemical plausibility of generated molecules.
• Identify limitations in current structure-based generative algorithms.
• Enhance the reliability of structure-based generation in drug discovery.

Main Methods:

• Developed two new metrics for chemical plausibility assessment.
• Analyzed property distributions of generated molecules.
• Compared generated molecules with high-throughput virtual screening hits for c-SRC kinase, Smoothened receptor, and dopamine D1 receptor.

Main Results:

• Structure-based generative algorithms can produce chemically implausible structures.
• Identified specific property distributions leading to implausible designs.
• Demonstrated differences between generated molecules and known drug-like molecules.

Conclusions:

• The new metrics provide valuable tools for evaluating generated molecules.
• Analysis enhances assessment of chemical plausibility and drug-likeness.
• Improves the application of structure-based generation in drug discovery pipelines.