Researchers introduce SciReasoner, a multimodal scientific foundation model that performs native structural reasoning across proteins, small molecules, and inorganic crystals by discretizing coordinates and topologies into a unified structure-aware vocabulary.
- In homology-controlled Gene Ontology prediction, it increases F_max from 0.42 to 0.55 for Cellular Component annotation of low-homology proteins.
- It raises single-step retrosynthesis accuracy from 0.63 to 0.72 while generating fragment-level disconnection and precursor-verification traces.
- Its representations separate elemental and compound phases and resolve high- and low-band-gap regimes in materials science.
- SciReasoner achieves state-of-the-art performance on 67 out of 86 benchmarks, with expert evaluation preferring its reasoning traces in 98% of cases.
By making structure an inspectable substrate for reasoning under scientific constraints, the model connects accurate prediction with interpretable scientific inference.