Monroe: MFM for In-Context Probabilistic Inference

chaumian · x · 2026-08-21

The paper presents Monroe, a new Molecular Foundation Model (MFM) for in-context probabilistic inference. Key innovations include pre-training on over 81 million molecules from the PM6 dataset, improved graph stereochemistry representation, conformer denoising losses, and the use of a prior-data-fitted model (TabPFN) for downstream tasks. Monroe matches or exceeds state-of-the-art on Polaris benchmarks and shows significant improvements on activity cliff benchmarks.

Original post →

More from Research

Research channel →