AI-driven computational drug design: tools, workflow and challenges

Area of Science:

• Computational chemistry and cheminformatics
• Artificial intelligence in drug discovery
• Medicinal chemistry and pharmacology

Background:

• Traditional drug discovery is lengthy, costly, and has high attrition rates (>90%).
• Computer-aided drug design faces limitations in scalability and biological system modeling.
• Artificial intelligence (AI) offers transformative potential to overcome these challenges.

Purpose of the Study:

• To review AI-based approaches across the drug discovery pipeline.
• To highlight emerging AI computational tools and their impact on medicinal chemistry.
• To provide a balanced overview of AI in drug discovery, including limitations and future directions.

Main Methods:

• Exploration of AI applications in target identification (e.g., graph-based models).
• Review of deep learning for virtual screening (e.g., GNINA, AtomNet) and de novo design (e.g., REINVENT, RANC).
• Discussion of AI-driven ADMET prediction (e.g., ADMETlab 2.0) and structural modeling (e.g., AlphaFold).

Main Results:

• AI platforms provide a comprehensive approach, shifting from narrow focus to integrated solutions.
• AI tools accelerate various stages: target discovery, virtual screening, molecular generation, lead optimization, and retro-synthesis.
• Recent advancements like AlphaFold significantly expand the availability of drug targets.

Conclusions:

• AI is reshaping medicinal chemistry, offering more robust and efficient drug discovery processes.
• Addressing dataset bias, reproducibility, and real-world applicability is crucial for clinical translation.
• AI holds promise for developing a more reproducible and clinically applicable drug discovery platform.