RosettaFold3

RosettaFold3 is a unified biomolecular structure-prediction system that handles proteins, nucleic acids, and small molecules within a single multimodal-transformer-plus-diffusion framework. It predicts 3D conformations of protein–ligand, protein–DNA, and protein–RNA complexes with atom-level conditioning, replacing what was previously a patchwork of specialized models with one architecture. The system supports complex assemblies and integrates structural priors across modalities, delivering high accuracy across diverse biomolecular systems with substantially less computational overhead than traditional pipelines.

RosettaFold3 marks a shift in computational structural biology from single-molecule prediction toward modeling whole biomolecular assemblies — the level at which most disease biology actually operates. The unified approach enables more reliable drug-discovery workflows, functional annotation of genetic variants, and mechanistic study of disease-associated complexes. Within Microsoft AI for Science, it is a concrete instance of the “fifth paradigm” — combining physics, data, and generative learning to compress experimental cycles in therapeutic development.