Impact of AI for drug development process

Source: The post Impact of AI for drug development process has been created, based on the article “The use of AI in drug development” published in “The Hindu” on 17th May 2024.

UPSC Syllabus Topic: GS Paper 3-Science and Technology- developments and their applications and effects in everyday life

Context: The article discusses how AI can speed up drug development. AI helps identify target proteins and predict drug interactions. Advanced AI tools like AlphaFold 3 improves accuracy. However, AI has limitations and requires significant computing infrastructure, which India currently lacks.

For detailed information on AlphaFold 3 read this article here

How does drug development start?

Drug development starts with identifying and validating a target, usually a protein.

Computers analyze target protein sequences to find the best-fitting drug from millions of small molecules. This process saves time and money by avoiding laboratory experiments.

Once a target protein and suitable drug are identified, the pre-clinical phase tests the drug’s safety and toxicity on cells and animals.

The clinical phase then tests the drug on human patients for efficacy and safety.

Finally, the drug undergoes regulatory approval, marketing, and post-market surveys.

Due to a high failure rate, the discovery phase limits the number of drugs that progress to the pre-clinical and clinical phases.

How can AI help the drug development process?

AI can speed up target discovery by cutting down time and increasing prediction accuracy.

Tools like AlphaFold and RoseTTAFold use deep neural networks to predict three-dimensional protein structures.

AlphaFold 3 and RoseTTAFold All-Atom can predict interactions for proteins, DNA, RNA, small molecules, and ions.

In a test of 400 drug-target interactions, AlphaFold 3 accurately predicted interactions 76% of the time, compared to 40% for RoseTTAFold All-Atom.

These AI tools save money and avoid time-consuming lab experiments.

What are the drawbacks?

1. AI tools can provide up to 80% accuracy in predicting interactions, which drops significantly for protein-RNA interactions.
2. AI tools only aid in target discovery and drug-target interaction phases, not in pre-clinical or clinical development phases.
3. Diffusion-based architectures in AI models can cause hallucinations, leading to incorrect predictions due to insufficient training data.
4. The code for AlphaFold 3 is not publicly available, limiting independent verification and broader use.
5. AI-derived molecules might not succeed in later drug development phases, despite initial promising predictions.

What about India?

1. India needs large-scale computing infrastructure with fast GPUs for developing AI tools in drug development. GPU chips are expensive and quickly become outdated.
2. India lacks skilled AI scientists compared to the U.S. and China.
3. Despite a rich history in protein X-ray crystallography and structural biology, India couldn’t establish a first-mover advantage in AI drug development.
4. India’s growing pharmaceutical industry can lead in applying AI tools for target discovery and drug testing.
5. Investment in computing infrastructure and training AI scientists is essential for progress.

Question for practice:

Discuss how AI can enhance and accelerate the drug development process.