Imagine a translator who doesn't learn a new language from scratch but instantly grasps the general meaning from a few examples. That's how Monroe works — a model for predicting molecular properties. Previously, such models required extensive tuning for each task, much like a translator learning a new language from zero. Monroe, on the other hand, looks at a few examples and immediately gives an answer.

This is crucial because two molecules with the same composition but different shapes can behave differently. Monroe distinguishes such cases.

For prediction, Monroe uses TabPFN — a model that compares new molecules with known ones. This allows it to work quickly and accurately, even with limited data. In tests, Monroe outperformed other models, especially in cases where small changes in a molecule significantly alter its activity.