An interpreter gets distracted by every rustle in the hall, and a one-hour meeting stretches into two. Similarly, molecular foundation models learn the language of molecules from vast corpora, but each new conversation requires fine-tuning. Monroe changes the rules: it doesn't learn a new language but instantly finds common meaning from a few examples.

Its key feature is an improved representation of stereochemistry: for E/Z double bonds and R/S chiral centers, auxiliary edges are added to distinguish isomers. This is like an interpreter who not only distinguishes words but also the gestures that change meaning.

Instead of fine-tuning, Monroe uses TabPFN, a model trained on synthetic tabular data. From molecular embeddings and their labels, it produces a probabilistic prediction for a test molecule in a single pass. This in-context probabilistic inference works like an experienced interpreter who grasps the essence of a conversation from a few remarks without flipping through a dictionary.

On 28 Polaris tasks, Monroe wins or is statistically indistinguishable from the best method in 100% of cases, with an average rank of 3.71. On 30 MoleculeACE tasks, specifically targeting "activity cliffs" (sharp activity changes with small structural differences), the model achieves the best overall RMSE of 0.629 and the best cliff RMSE of 0.737.

Moreover, applying TabPFN to existing models MiniMol and CheMeleon yields new strong variants, MiniMolPFN and CheMeleonPFN, confirming the approach's versatility.

Monroe demonstrates that in-context probabilistic inference can replace fine-tuning, reducing computational costs and improving performance on small data. This paves the way for more flexible and faster methods in cheminformatics. The interpreter no longer needs to learn every dialect, just understand the common language.