AI SCIENCE
AI is helping chemists plan new molecules
Designing new molecules is one of the hardest parts of chemistry.
Whether researchers are making a new drug or a new material, they need to plan every reaction carefully.
That means working out which steps are most likely to succeed and which ones could waste time.
A key challenge is retrosynthesis, where chemists start with the final molecule and work backwards to find the best way to make it.
Computers can search through huge numbers of possible routes, but they still struggle with the kind of judgement experienced chemists bring.
Researchers at EPFL have developed a new AI framework called Synthegy to help with this.
Instead of generating molecules on its own, it helps existing chemistry software judge possible routes.
The main bits:
Synthegy lets chemists guide AI tools using plain language.
It helps rank possible reaction routes and mechanisms.
The goal is to support chemists, not replace their expertise.
Plain English enters the lab
Chemists can give Synthegy simple written instructions, such as asking it to form a ring early or avoid unnecessary protection steps.
The system then reviews possible reaction paths, scores how well they match the request, and explains its reasoning.
Synthegy can also help analyse reaction mechanisms, which show how reactions happen step by step.
It checks possible pathways and helps point researchers towards the ones that make the most chemical sense.
In testing, 36 chemists reviewed its results across 368 valid evaluations. Their views matched Synthegy’s results 71.2% of the time on average.
Retrosynthesis sounds like a villain origin story, and honestly, fair - MG