Use of AI in Drug Development and its Drawbacks

Context

• The drug development process is traditionally expensive and time-consuming.
• Artificial Intelligence (AI) has introduced new possibilities to accelerate this process.

Drug Development Process

• Identifying and Validating a Target: The process begins by identifying and validating a biological molecule (target), usually a protein or gene, which the drug will bind to.
  • Druggable proteins have specific sites where drugs can dock to exert their effects.
• Discovery Phase: Target proteins are identified, and their sequences are fed into a computer.
  • The computer searches for the best-fitting drug from millions of stored small molecule structures.
  • This computational process bypasses time-consuming and costly laboratory experiments, reducing failure rates.
  • Once a suitable protein target and drug are identified, the research moves to the pre-clinical phase.
• Pre-Clinical Phase: Potential drug candidates are tested on cells and animals for safety and toxicity.
• Clinical Phase: The drug undergoes trials on a small number of human patients for initial testing, followed by more extensive patient trials for efficacy and safety.
• Regulatory Approval and Post-Market Surveillance: The drug is reviewed for regulatory approval and, upon approval, is marketed and monitored for any post-market issues.

Role of AI in Drug Development

• Accelerating Target Discovery and Drug-Target Interaction:
  • AI can significantly reduce the time and cost involved in drug development and increase the accuracy of predictions.
  • AI-based prediction tools like AlphaFold and RoseTTAFold have made significant advancements in computational drug development.
• AlphaFold and RoseTTAFold:
  • Developed by DeepMind (Google) and the University of Washington, respectively.
  • These tools use deep neural networks to predict the three-dimensional structures of proteins.
  • New versions, AlphaFold 3 and RoseTTAFold All-Atom predict interactions for proteins, DNA, RNA, and small molecules with higher accuracy.
  • In tests, AlphaFold 3 predicted interactions accurately 76% of the time, compared to 40% by RoseTTAFold All-Atom.

Drawbacks of AI in Drug Development

• Accuracy Limitations: Current Artificial Intelligence (AI) tools can achieve up to 80% accuracy in predictions, which drops significantly for protein-RNA interactions.
• Limited to Early Phases: AI tools primarily assist in the target discovery and drug-target interaction phases, but pre-clinical and clinical phases still need to be completed without any guaranteed success.
• Model Hallucinations: Diffusion-based architectures can produce incorrect predictions due to insufficient training data.
• Code Availability: The code for AlphaFold 3 has not been released, limiting independent verification and broader utilisation.

Challenges and Opportunities in India

• Infrastructure Needs: Developing AI tools requires large-scale computing infrastructure with advanced GPU chips, which are expensive and have quick expiration dates.
• Skill Gap: India lacks the number of skilled AI scientists compared to the U.S. and China, limiting its ability to develop first-mover AI tools for drug development.
• Potential in Pharmaceutical Applications: Despite challenges, India’s growing pharmaceutical sector has the potential to lead in applying AI tools for target discovery, identification, and drug testing.

Sharing is caring!