Computational drug discovery is the search for a new medicine's molecule on a computer, before anything is made in a laboratory. It is also called computer-aided drug discovery, or in silico drug discovery. In silico means "done on a computer". The active ingredient of a medicine is one molecule. Finding it means choosing one candidate from millions, or even billions, of molecules that might have worked.
The search starts from a target. A target is a protein in the body that a disease depends on. If a molecule sticks to the right spot on that protein, it can switch the protein off, and the disease slows down. The molecule has to fit the protein the way a key fits a lock.
Here is a typical case. A discovery team at a drug company in Bengaluru wants a new medicine for psoriasis, a long-lasting skin disease. First, the team studies gene data from patients' skin on a bioinformatics data platform. The data shows which protein is far more active in the patches of diseased skin. That protein becomes the target. This use of gene data is explained in genomics.
Next, the team searches for molecules that fit the target. Molecular modelling software tests millions of existing molecules against a 3D model of the protein. AI drug discovery models go a step further. They design new molecules that do not exist yet. Many of these models are built on foundation models, very large AI models trained on huge amounts of data.
Then the team asks what each good candidate would do inside a person. In silico drug modelling simulates the body on a computer. It predicts how much of the medicine is absorbed, how long it stays in the blood and how it is removed. Candidates that would be cleared too fast, or that could harm the liver, are dropped early.
Finally, chemists plan the cream or tablet that will carry the molecule. This recipe is called the formulation. AI for chemistry and materials R&D predicts which recipes will work, so that the chemists test fewer of them by hand.
Why do all this on a computer? Making and testing one molecule at the bench takes days and costs money. A computer can check millions of molecules in a few days. So only a few hundred molecules are actually made and tested, on the instruments of the research laboratory. Some laboratories now connect their instruments to an AI model that chooses each next experiment. This is called a self-driving lab.
The computer does not replace testing. The best few molecules are still tested in cells, then in animals, and finally in people, in a clinical trial. Most candidates still fail along the way. The open question is whether molecules designed by AI will fail less often than molecules found the old way.