MIT Scientist Teaches AI Chemistry to Design New Drugs

Connor Coley of the Massachusetts Institute of Technology works at the intersection of chemistry and machine learning. His new approach helps AI systems do more than find patterns in data: they also grasp the fundamental principles of chemistry and use that knowledge to design new drugs.

Why conventional ML falls short

Traditional machine learning models are trained on huge volumes of data and look for patterns. In chemistry, that is not enough: a molecule that looks statistically similar can have entirely different properties. The AI needs to understand real chemical rules, such as how atoms interact and why electrons distribute themselves in a particular way.

That is why Coley built chemical principles into his models. The AI no longer guesses; it reasons, like an experienced chemist.

How it works

Coley's approach uses so-called physics-informed neural networks. They combine the power of deep learning with explicit constraints drawn from chemistry. The model can propose a molecule that:

• Has never appeared in the training set
• But obeys chemical laws, violating neither valence nor electron balance
• And probably has the properties a drug needs

It is like giving the AI not just a set of examples but a chemistry textbook, then asking it to solve a creative problem.

Applications and potential

"We want the AI not just to predict which molecules will work, but to be able to explain why," says

Coley.

The approach is already showing results. The AI finds drug candidates that human chemists would have passed over because they are strange or atypical, yet still functional. This speeds up the early stages of development: instead of synthesizing hundreds of compounds, researchers can first filter down to the 10–20 best options. For the pharmaceutical industry this is critical: developing a single drug costs billions and takes years.

What this means

It signals a shift in how machine learning is evolving. The first generation of AI was purely statistical, guessing at patterns. The second generation builds in expert knowledge and physical laws. AI becomes not just a search tool but a researcher that reasons by rules. For chemistry and biology, this may mean faster progress not only in drug development but also in the discovery of new materials and catalysts.