ARTIFICIAL INTELLIGENCE IN DRUG DEVELOPMENT

The process of developing a drug begins with identifying and validating a target. A target is typically a biological molecule, such as a gene or a protein, to which a drug binds to exert its effect. Most targets are proteins, and only those with ideal docking sites for drug molecules are considered "druggable" proteins.

• In the discovery phase, target proteins are identified, and their sequences are fed into a computer. The computer then searches for the best-fitting drug from a library of millions of small molecules, whose structures are stored in the system. This process relies on known structures of both the target protein and the drug. If these structures are unknown, the computer uses models to predict potential binding sites. This computational approach bypasses the need for time-consuming and costly laboratory experiments that often have a high failure rate.
• Once a suitable protein target and its corresponding drug are identified, the research progresses to the pre-clinical phase. During this stage, potential drug candidates are tested outside of biological systems, typically using cells and animals, to evaluate the drug's safety and toxicity.
• Following successful pre-clinical testing, the drug enters the clinical phase. Initially, it is tested on a small group of human patients to assess its safety. If successful, the testing is expanded to a larger group of patients to evaluate both efficacy and safety.
• After passing clinical trials, the drug undergoes regulatory approval. Once approved, the drug is marketed and continues to be monitored through post-market surveillance to ensure ongoing safety and efficacy.

Artificial Intelligence (AI) has the potential to revolutionize target discovery and drug-target interaction analysis by significantly reducing the time required, increasing prediction accuracy, and saving costs. Recent advancements in AI-based prediction tools, such as AlphaFold and RoseTTAFold, represent breakthroughs in computational drug development.

AlphaFold and RoseTTAFold: A Scientific Breakthrough

Developed by researchers at DeepMind, a Google company, and the University of Washington, U.S., respectively, AlphaFold and RoseTTAFold are based on deep neural networks. These networks utilize vast amounts of input data to generate three-dimensional structures of proteins. In recent years, the upgraded versions of these tools, AlphaFold 3 and RoseTTAFold All-Atom, have further advanced this field.

Capabilities of Upgraded AI Tools

AlphaFold 3, developed jointly by Isomorphic Labs (a DeepMind spinoff), and RoseTTAFold All-Atom, have significantly enhanced capabilities compared to their predecessors. These new versions can predict not only static structures of proteins and protein-protein interactions but also structures and interactions involving any combination of proteins, DNA, RNA, modifications, small molecules, and ions. They employ generative diffusion-based architectures, a type of AI model, to predict structural complexes.

Improved Prediction Accuracy

In a test involving 400 interactions between targets and their small molecule drugs, AlphaFold 3 accurately predicted interactions 76% of the time, compared to 40% accuracy by RoseTTAFold All-Atom. This marked improvement highlights the potential of these AI tools to streamline the drug development process by enhancing the accuracy of drug-target interaction predictions.

4. Drawbacks of AI in Drug Development

Despite the promise and potential AI brings to drug development, there are several limitations and challenges associated with its use.

Limited Accuracy

AI tools, while advanced, can only achieve up to 80% accuracy in predicting drug-target interactions. The accuracy significantly decreases for protein-RNA interaction predictions, highlighting a limitation in the scope of reliable predictions.

Restricted to Early Phases

These AI tools are primarily beneficial for the initial phase of drug development—target discovery and drug-target interaction. They do not eliminate the need for the pre-clinical and clinical development phases, where there is no assurance that AI-derived molecules will succeed.

Model Hallucinations

Diffusion-based architectures, employed by these AI tools, are susceptible to model hallucinations. This issue arises when there is insufficient training data, leading the tool to generate incorrect or non-existent predictions.

Limited Accessibility

Unlike earlier versions of AlphaFold, DeepMind has not released the code for AlphaFold 3. This restriction limits independent verification, broad utilization, and its application for studying protein-small molecule interactions. This lack of transparency and accessibility is a significant drawback for the scientific community.

5. India's AI Landscape in Drug Discovery

Developing advanced AI tools for drug development requires substantial computing infrastructure, particularly high-speed Graphics Processing Units (GPUs) capable of running multiple complex tasks involving long sequences. These GPU chips are expensive and rapidly become obsolete as newer, faster models are released annually.

Infrastructure and Expertise Challenges

India currently lacks the necessary large-scale computing infrastructure to support the development of AI tools in drug development. Additionally, there is a shortage of skilled AI scientists in India, unlike in countries like the U.S. and China. This shortage has hindered Indian researchers from establishing a first-mover advantage in developing AI tools for drug development, despite India's rich history in protein X-ray crystallography, modelling, and other structural biology fields.

Opportunities for Growth

However, India has a growing number of pharmaceutical organizations, which positions the country well to become a leader in applying AI tools for target discovery, identification, and drug testing. By investing in advanced computing infrastructure and cultivating AI expertise, India can leverage its strong foundation in structural biology to make significant contributions to AI-driven drug development.

6. Way Forward

By addressing the critical areas, India can transform its role from a spectator to a frontrunner in the exciting field of AI-driven drug discovery. This will not only benefit the country's own healthcare sector but also contribute to global efforts in developing life-saving medications faster and more efficiently.