BioEmu Microsoft

Our verdict

BioEmu is Microsoft Research’s open-source model for sampling protein structural ensembles directly from sequence. Unlike AlphaFold2 or ESMFold, which return one or a handful of static structures, BioEmu targets the equilibrium ensemble and even reports relative free energy differences. In practice, it generates thousands of statistically independent samples per hour on a single GPU and shows ~1 kcal/mol free-energy errors versus MD and experiment, which is compelling for hypothesis generation. The Linux-only pip package is straightforward to install, with a CLI and Python API, and it bundles MSA retrieval via ColabFold/AlphaFold2. There’s helpful steering to reduce clashes and an optional HPacker step for side-chain reconstruction. Limitations are clear: monomers only, no explicit ligands or membranes, and no official hosted API or enterprise support. It also relies on MSAs, so it lacks ESMFold’s single-sequence convenience. Still, for academics and R&D teams studying dynamics or cryptic pockets, the price-to-capability ratio is excellent—especially given the MIT license and optional Azure AI Foundry path for cloud deployment.