bioemu2 README

The project's README from github.com/microsoft/bioemu2. As one markdown file: README.md.

BioEmu2 logo

BioEmu2

A foundation model for biomolecular structure and dynamics

Predict folded structures, sample equilibrium ensembles, and propagate structures forward in time.

Installation · Quick start · Tutorials · Documentation


BioEmu2 simulating the PUMA BH3 peptide binding MCL-1

BioEmu2 simulating the PUMA BH3 peptide binding MCL-1.

BioEmu2, short for "biomolecular emulator", is developed by Microsoft Research AI for Science.

Key features include:

System requirements

BioEmu2 requires Linux and an NVIDIA GPU; macOS and Windows are not supported. The cuda extra installs CUDA 13 cuEquivariance, which requires compute capability 7.5 or newer, such as NVIDIA T4, A100, RTX A6000, L40S, H100, and B200 (CUDA 13 release notes). GPUs with compute capability 7.0 (Volta, such as NVIDIA V100) need the CUDA 12 installation instead. We have tested V100 (compute capability 7.0), T4 (7.5, with the CUDA 12 installation), A100 and RTX A6000 (8.0 and 8.6), and B200 (10.0); other eligible GPUs have not been verified end to end. Check your GPU's compute capability with nvidia-smi --query-gpu=name,compute_cap --format=csv.

Context-parallel sampling requires at least four GPUs.

Installation

Use Python 3.12, 3.13, or 3.14 on a supported Linux machine. We recommend installing BioEmu2 in an isolated virtual environment or conda environment. To install in an isolated virtual environment, create and activate it from the repository root:

python3.12 -m venv .venv
source .venv/bin/activate

Replace python3.12 with the name or path of your Python 3.12, 3.13, or 3.14 executable if needed. Use python -m pip so packages are installed into the active Python environment rather than a different interpreter's pip.

Run the installation command for your GPU from the repository root. For GPUs with compute capability 7.5 or newer, such as NVIDIA T4, A100, RTX A6000, L40S, H100, and B200, use the cuda extra:

python -m pip install ".[cuda]"

Compute capability 7.0 (Volta, such as NVIDIA V100) is the oldest supported; on these GPUs, use the cuda12 extra with the PyTorch CUDA 12.6 package index:

python -m pip install ".[cuda12]" \
  --extra-index-url https://download.pytorch.org/whl/cu126

GPUs with compute capability below 7.0 are not supported.

We tested BioEmu2 and evaluated its benchmark metrics with Python 3.12. Python 3.13 and 3.14 pass the test suite, but we have not evaluated benchmark metrics with them.

For development, see DEVELOPER.md.

For Jupyter or VS Code notebooks, follow the notebook quickstart to install visualization dependencies and register a kernel.

Quick start

CLI

Sample eight equilibrium conformations of adenylate kinase (PDB 1AKE):

bioemu2 sample-equilibrium \
  --pdb-id 1AKE \
  --out outputs/adk

Or predict a folded structure from a PDB entry:

bioemu2 predict-structure \
  --pdb-id 1UBQ \
  --num-samples 1 \
  --out outputs/ubiquitin

Propagate each sampled adenylate kinase conformation by 1 ns:

bioemu2 propagate \
  --x0-cif outputs/adk/topology.cif \
  --x0-xtc outputs/adk/samples.xtc \
  --lag-time-ns 1 \
  --out outputs/adk-propagated

Run bioemu2 --help to list all commands.

Python API using EmuSession

Construct an EmuSession to sample and propagate structures:

from pathlib import Path

from bioemu2 import EmuSession, System, load_model, save_frames

model = load_model()

# Predict a folded structure for ubiquitin.
ubiquitin = System.from_pdb_id("1UBQ")
prediction = EmuSession(ubiquitin, model=model, mode="structure").sample(num_samples=1)
save_frames(prediction, Path("outputs/ubiquitin"))

# Sample equilibrium conformations of adenylate kinase.
adk = System.from_pdb_id("1AKE")
emu = EmuSession(adk, model=model, mode="equilibrium")
samples = emu.sample(num_samples=8)
save_frames(samples, Path("outputs/adk"))

# Propagate each adenylate kinase conformation by 1 ns.
endpoints = emu.propagate(samples, lag_time_ns=1.0)
save_frames(endpoints, Path("outputs/adk-propagated"))

save_frames() writes samples.xtc and topology.cif to each output directory.

See define molecular systems, predict and sample structures, and propagate structures for more options. The functional Python APIs provide high-level sampling and propagation functions similar to the CLI commands, plus lower-level workflows for explicit control of embeddings and samplers.

More workflows

Steer sampling

Choose a steering sampler to favour structures with desired properties:

bioemu2 sample-equilibrium \
  --system src/bioemu2/assets/examples/systems/myoglobin_101m.yaml \
  --out outputs/myoglobin \
  --sampler edm_gd

See steer sampling and define steering potentials to select an algorithm and configure its objective.

Sample large structures across multiple GPUs

Context parallelism distributes the model's pair representations across a square mesh of GPUs, reducing the memory required on each device:

CUDA_VISIBLE_DEVICES=0,1,2,3 \
torchrun --standalone --nproc-per-node=4 \
  --module bioemu2.cp sample-equilibrium \
  --pdb-id 1AKE \
  --checkpoint bioemu2-preview \
  --num-samples 3 \
  --out outputs/adk-cp

See sample large structures with context parallelism for mesh requirements, the Python API, and current limitations.

Inspect and subsample MSAs

Retrieve an alignment from ColabFold or load your own A3M files, inspect the raw sequences, and select or randomly subsample rows with Python indexing. Pass the resulting MSA directly to EmuSession to control the alignment used for structure prediction, equilibrium sampling, or propagation without manually constructing embeddings.

See use multiple sequence alignments for local-file loading, MSA subsampling, and custom-MSA inference.

Tutorials

Notebook quickstart

The examples directory contains complete workflows as regular Python files split into executable cells with # %% markers. This Python Interactive format provides notebook-style cells while keeping the examples as plain .py files. They require a Linux machine with a supported NVIDIA GPU and network access.

Activate a Python environment as described in installation. From the repository root, install BioEmu2 with the visualization dependencies and register a kernel. The viz extra includes ipykernel and py3Dmol:

python -m pip install ".[cuda,viz]"
python -m ipykernel install --user --name bioemu2 --display-name "BioEmu2"

On compute capability 7.0 GPUs, such as NVIDIA V100, replace the first command with:

python -m pip install ".[cuda12,viz]" \
  --extra-index-url https://download.pytorch.org/whl/cu126

Select the BioEmu2 kernel in Jupyter or VS Code. If registration succeeds but the kernel does not appear in VS Code, restart VS Code and try again. To check that it uses the environment you installed into, run import sys; print(sys.executable) in a cell. The examples load system YAML from the installed bioemu2 package, regardless of the working directory. Relative outputs/ paths are written under the current working directory.

Open a single-GPU example in VS Code and select Run Cell, or run Jupyter: Run Current File in Interactive Window from the Command Palette. Launch the context-parallel example with torchrun as described in the context-parallel sampling guide.

To run without VS Code, convert an example with Jupytext and open it in JupyterLab:

python -m pip install jupytext jupyterlab
jupytext --to notebook examples/sample_structures.py
jupyter lab examples/sample_structures.ipynb

Select the BioEmu2 kernel, then run individual cells or choose Run All Cells from the Run menu.

In VS Code, you can instead run Jupyter: Export Current Python File as Jupyter Notebook from the Command Palette. Generated .ipynb files are ignored by Git.

License

BioEmu2 is released under the MIT License. It includes code and reference data adapted from other open-source projects; the third-party notices identify those components and their licenses.

bioemu2 (microsoft/bioemu2)

drug-discovery
Searching Database…

422 ·GitHub
Bioemu2
What is this?BioEmu2 (biomolecular emulator) is a foundation model from Microsoft Research AI for Science that predicts folded protein/nucleic-acid structures, samples equilibrium structural ensembles, and propagates structures forward in time.
Is this a Python library, a CLI tool, or both?Both: you install it as a Python package and can either run it from the command line (e.g. bioemu2 sample) or call it programmatically via a Python API built around an EmuSession object.
How do I install and run it?Create a Python 3.12/3.13/3.14 virtual environment on Linux, then run the pip install command with the matching GPU extra (cuda or cuda12) from the repository root, and use the bioemu2 CLI or EmuSession to sample structures.
What hardware do I need to run this?A Linux machine with an NVIDIA GPU of compute capability 7.0 or higher (e.g. V100, T4, A100, H100, B200); GPUs below 7.0 are not supported, and context-parallel sampling needs at least four GPUs.
Does it run on macOS or Windows, or on any NVIDIA GPU?No - macOS and Windows aren't supported at all, and GPUs with compute capability below 7.0 can't run it; older Volta cards like the V100 need a separate CUDA 12 install path instead of the default CUDA 13 build.
Who is this built for?Computational biologists and researchers who need to predict protein structures, sample conformational ensembles, or run molecular dynamics-style propagation without classical physics-based simulation.
Can this run on a laptop GPU or only data-center cards?It needs a discrete NVIDIA GPU with compute capability 7.5+ (T4, A100, RTX A6000, L40S, H100, B200) for the standard CUDA 13 install, or 7.0 (V100) via the CUDA 12 path; consumer laptop GPUs without sufficient compute capability aren't supported.
What is an EmuSession?It is the main Python API object used to construct molecular systems and then sample, propagate, or steer their structures with a few method calls instead of using the CLI.
What does 'steering sampling' let me do?You can pick a steering sampler and configure potentials so BioEmu2 biases its generated structures toward desired properties, rather than just sampling the unconditioned equilibrium ensemble.
How does it handle structures too large for one GPU?Context parallelism splits the model's pair representations across a square mesh of multiple GPUs (minimum four), letting you sample larger structures than a single GPU's memory would allow.
Can I use my own sequence alignments?Yes - you can pull alignments from ColabFold or load your own A3M files, inspect and subsample the sequences, and feed the resulting MSA directly into an EmuSession to control structure prediction or sampling.
What are similar projects?RFD4-Proteina, alphafold, SuperMimic-Reproduced, EasyWAM, SolarWM, dreamforge, damo-radar, VisionHOPE
Which topics is it in?computational-biology, drug-discovery, science, foundation-models, jupyter-notebook, simulation, gpu-acceleration, academic-research, distributed-systems, nvidia, cuda, pytorch
Creator
microsoft
Size
6.5 MB
Stars
422
Updated
CommitOct 23Releasereadme-mediaOct 19
Age
2mo
Commits · 90 Days