Molecular modelling software is software with which chemists build 3D models of molecules on a computer, and predict how those molecules behave. Drug companies use it mainly to search very large collections of molecules for one that fits a disease's target protein. A target protein is a protein in the body that a disease depends on. It is also called computational chemistry software or molecular design software.
The main job is called virtual screening. The software takes a 3D model of the target protein. It finds the pocket on the protein where a molecule could stick. Then it tries each molecule from a library in that pocket, in thousands of positions and twists. This fitting is called docking. For each molecule, the software calculates a score: an estimate of how tightly the molecule would stick. The tighter it sticks, the better it may block the protein.
Take, for example, a medicinal chemist at a drug company in Hyderabad. She is looking for a molecule that blocks a protein involved in rheumatoid arthritis. Her company has the protein's 3D shape, measured in an earlier experiment. She docks a library of 5 million molecules that suppliers can make to order. The job is spread over a thousand processor cores by a distributed computing framework, and it finishes in about two days. The software ranks every molecule by its score. She looks closely at the top 2,000. She removes molecules that would be hard to make or that resemble known poisons. Then she orders the best 100. A laboratory tests them, and 6 of them actually stick to the protein. Six real starting points from 100 tests is a good result. Testing all 5 million at the bench would be impossible.
The software does more than screening. It simulates how the protein and the molecule move together over time, which is called molecular dynamics. It helps chemists change a molecule one atom at a time and predict whether each change improves the fit. It also predicts properties such as how well the molecule dissolves in water.
The results come back to the laboratory. There, chromatography and mass spectrometry confirm what is actually in each tube. The software also uses gene and protein data from a bioinformatics data platform.
How is molecular modelling different from AI drug discovery? Classic molecular modelling uses the laws of physics and chemistry to calculate how two molecules interact. AI drug discovery learns patterns from data, and it can also invent molecules that are in no library. Many teams now use both: the AI proposes molecules, and the physics-based software checks how well they fit. The same kind of software is used outside medicine too, for example to design new materials, as described in AI for chemistry and materials R&D.