PURE: An AI Tool Makes Designing Safer, More Effective Medicines Easier
An AI tool developed by IIT Madras has made discovery of highly safe, effective drug analogues easily possible. In the case of the FDA-approved cancer drug Sorafenib, starting from its original structure, PURE produced hundreds of thousands of possible analogues, and then narrowed them down using basic drug-likeness filters. A manageable group of candidates that were selected exhibited docking scores comparable to, and in some cases, superior to those of Sorafenib
Finding a new medicine is often compared to searching for a needle in a haystack, but in reality, the haystack is unimaginably vast. Chemists estimate that the total number of drug-like molecules that could theoretically exist is around 1060. This figure is so large that even if every laboratory in the world synthesised new molecules continuously for centuries, they would still explore only a tiny fraction of this chemical space. Traditional trial-and-error approaches, while scientifically rigorous, simply cannot scale to the magnitude of possibilities nature allows.
This is where intelligent computational tools can make a meaningful difference. By helping researchers decide which molecules are worth investigating, these systems can dramatically narrow the search space, making discovery quicker, more systematic, and more affordable.
To support chemists navigating this landscape, we have developed Policy-guided Unbiased REpresentations (PURE), an AI-driven system (Journal of Cheminformatics) that builds on the foundation of established reaction chemistries. Unlike earlier generative models that depend on hard-coded or manually curated rules, PURE learns the types of structural changes that real chemical reactions can produce. It does this by analysing reaction data and identifying transformation patterns, effectively absorbing a practical sense ofchemical feasibility. When PURE proposes a new molecule, it does not generate chemically unrealistic structures; it suggests ones that a synthetic chemist could possibly construct in the laboratory.
This reaction-aware approach gives PURE an important advantage: it can recommend a rich variety of modifications around any starting molecule, but always within the boundaries of real-world chemistry. The result is a broader, yet still practical, set of candidate structures for early-stage drug discovery.
A faster way to improve existing medicines
Most medicines begin as an initial “lead compound” that shows some promise against a biological target. From there, chemists refine the structure — a painstaking process known as lead optimisation. They adjust functional groups, tweak side chains, or replace ring structures, all with the aim of improving potency, reducing side effects, or overcoming resistance. This iterative decision-making process heavily relies on expert intuition developed over years of training, and PURE aims to support this aspect of the workflow.
Given a known drug molecule, PURE can generate hundreds or thousands of structurally related analogues in minutes. These analogues typically maintain the essential molecular framework responsible for therapeutic activity, while exploring new regions of chemical space around it. This allows researchers to rapidly consider pathways they might not have thought to pursue manually.
A real-world example: Variants of Sorafenib
In one case study, PURE was used to generate new variants of sorafenib, an FDA-approved cancer drug. Starting from the original structure, PURE produced hundreds of thousands of possible analogues, and then narrowed them down using basic drug-likeness filters. From this filtered set, we selected a manageable group of candidates and conducted molecular docking simulations to assess the potential interaction of these molecules with relevant protein targets. Several of these AI-generated structures exhibited docking scores comparable to, and in some cases, superior to those of sorafenib, suggesting that PURE can identify synthetically feasible drug analogues that have high binding affinity, specific interactions, and favourable druglike properties worthy of further experimental study.
While docking results cannot confirm real-world effectiveness, they do provide a useful prioritisation signal. By starting with molecules that are synthetically plausible and computationally promising, chemists can potentially save months of time and considerable experimental resources.
PURE does not guarantee success; no AI tool can. However, it can help make the earliest stages of drug discovery more targeted and experimentally grounded.
Why such tools matter
In addition to aiding scientists in designing new and more effective drugs, an AI tool is particularly useful for the following three reasons:
Countering drug resistance: Diseases such as cancer, tuberculosis, and malaria frequently evolve resistance to existing treatments. One common strategy to overcome resistance is to design close structural analogues of successful drugs; molecules that work similarly but evade the new defence mechanisms of the disease. PURE’s ability to quickly generate synthetically feasible analogues makes it well-suited for this purpose.
Reducing harmful side-effects: Some highly effective drugs fail to reach their full potential because of toxicity or adverse reactions. By proposing alternate structures that preserve the drug’s therapeutic “core” while altering peripheral groups, PURE may help researchers identify safer variants. Even incremental improvements in safety can significantly affect patient outcomes, especially for drugs taken chronically.
Lowering the cost of early discovery: One of the biggest barriers in pharmaceutical research is cost. Synthesising and testing thousands of molecules can be prohibitively expensive, particularly for small labs, academic groups, or non-profit organisations. Tools like PURE help focus efforts on a manageable subset of promising candidates, making early discovery more accessible and potentially accelerating progress in underfunded disease areas.
AI cannot replace the laboratory testing
Despite their growing capabilities, AI systems like PURE cannot replace experimental science. Every AI-proposed molecule must still be synthesised, tested for toxicity and stability, evaluated in biological assays, and validated across multiple models before it can even begin to approach clinical relevance. The journey from a computer-generated structure to an approved medicine remains long, complex, and fundamentally empirical.
What PURE offers is a better starting point. Instead of searching blindly or relying solely on intuition, chemists can begin with a diverse, chemically grounded set of molecules selected to maximise the chances of success.
A tool for a global challenge
As cancers evolve and infectious diseases adapt, the need for efficient and scalable drug-discovery pipelines is more urgent than ever. AI is not a magic solution, but it can be a powerful ally. By expanding and refining the space of possibilities that researchers can realistically explore, systems like PURE help bridge the gap between what is theoretically imaginable and what is experimentally achievable. If tools like PURE continue to advance, they could make early-stage drug design faster, more affordable, and better aligned with real-world chemistry.
