Artificial intelligence in drug discovery, explained

Why discovering medicines is so hard

Bringing a new medicine to patients is one of the hardest challenges in science. Researchers must first understand a disease well enough to find a biological target — often a protein — that a drug could act on. They then search among vast numbers of possible molecules for one that binds the target, works safely in the body, and can be made into a stable medicine. The overwhelming majority of candidates fail during testing, and the whole process can take well over ten years and huge investment. Because there are more possible drug-like molecules than there are atoms in the solar system, exploring them one by one in the laboratory is impossible — which is exactly where computers can help.

How AI is used across the pipeline

AI is being applied at nearly every step of drug discovery. Machine-learning models sift through mountains of biological data to suggest promising targets for a disease. Other models predict how strongly a molecule will bind to a target, screening millions of virtual compounds far faster than laboratory testing. So-called generative models can even design entirely new molecules with desired properties, rather than only picking from existing ones. AI also helps predict whether a candidate might be toxic or poorly absorbed, so weaker options can be dropped earlier. Later, it can help identify suitable patients for clinical trials and analyse trial data. The aim throughout is to fail faster and cheaper, and to focus laboratory effort where it counts.

A landmark: predicting protein shapes

One of the most celebrated advances is the ability of AI to predict the three-dimensional shape of proteins from their building blocks. A protein's function depends on how it folds into a complex shape, and for decades working out these shapes required slow, expensive laboratory experiments. AI systems can now predict many protein structures with remarkable accuracy in a fraction of the time, and open databases have made huge numbers of predicted structures freely available to researchers worldwide. This matters for drug discovery because knowing a target's shape helps scientists design molecules that fit it precisely — although a predicted structure is a starting point that still needs laboratory confirmation.

What AI can and cannot do

It is important to be realistic about AI's role. AI is a powerful tool for generating and prioritising ideas, but it does not replace laboratory experiments or clinical trials. A model can only be as good as the data it learns from, and gaps or biases in that data can lead it astray; biology is also full of surprises that no computer can fully anticipate. A molecule that looks perfect on screen may behave differently in living cells or in the human body. AI can shorten the early stages and improve the odds, but a candidate medicine must still be tested rigorously for safety and effectiveness in people before it can be approved.

What this means for patients

For patients, AI in drug discovery is a reason for cautious optimism rather than instant transformation. Faster, cheaper early research could mean more candidate treatments reaching trials, more attention to rare diseases that were previously uneconomic to pursue, and potentially more tailored medicines. Several AI-designed molecules are already being tested in clinical trials, though it will take years to know how many succeed. Regulators such as the MHRA are actively developing frameworks to evaluate AI-supported research and ensure safety standards are upheld. The essential safeguards — independent evidence, careful trials and regulatory approval — remain firmly in place, protecting patients while the technology matures.