Skip to main content
Pre-release

This release is a pre-release and may not be stable for production use.

Proxide

Alpha — experimental, no API stability guarantees. This crate is part of an active research project; APIs may change without notice between versions. Not recommended for production use.

Proxide is a high-performance library for Protein I/O and Physics bridging in JAX. It combines a flexible Python/JAX frontend with a highly optimized Rust backend (_proxider) to provide fast structure parsing, force field parameterization, and seamless integration with JAX MD.

NOTE: This is a research library in active development.


🚀 Features

  • Hybrid Architecture:
    • Rust Backend: 25x faster parsing (PDB/mmCIF), 50x faster topology generation, and robust force field parameterization.
    • JAX Frontend: Differentiable physics, geometric deep learning utilities, and seamless GPU integration.
  • Robust I/O: Load PDB, mmCIF, and PQR files with automatic error handling and corrections.
  • Molecular Dynamics: Parse OpenMM XML force fields, assign GAFF parameters, and generate fully parameterized AtomicSystem objects for JAX MD.
  • Trajectory Support: High-performance parsing of XTC, DCD, and TRR trajectories.

📦 Installation

Proxide requires a Rust toolchain to build the backend extension.

Prerequisites

  • Python: 3.11+
  • Rust: 1.75+ (Install via rustup.rs)
  • C++ Compiler: For compiling HDF5/chemfiles dependencies if needed.

From Source

# Clone the repository
git clone https://github.com/maraxen/proxide.git
cd proxide

# Install with uv (recommended) or pip
uv pip install .

# For development (includes test dependencies)
uv pip install -e ".[dev]"

The installation process will automatically compile the Rust _proxider extension using maturin.

Optional: Espaloma Charge (ML partial charges)

Inference uses the JAX/Equinox port expaloma (bundled weights; no PyTorch/DGL at runtime). The [espaloma] extra pulls expaloma from PyPI (and rdkit).

uv pip install -e ".[espaloma]"

For local development against an editable expaloma checkout, use a path override (e.g. uv.sources in your workspace) or:

uv pip install -e ../expaloma
uv pip install -e ".[dev]"

OpenFF toolkit wrapper (legacy upstream espaloma_charge): uv pip install -e ".[espaloma-openff]".

API: proxide.chem.partial_charges (assign_espaloma_charges_rdkit, etc.). Validation uses golden vectors under tests/data/espaloma_golden/; refresh those files when bumping expaloma.

pytest -m espaloma

🛠️ Usage

Loading a Structure

Use the high-level parse_structure function for fast, robust parsing:

from proxide import parse_structure

# Load a PDB file to a unified Protein object
protein = parse_structure("path/to/structure.pdb")

# Access data as JAX arrays
print(protein.coordinates.shape)  # (N_residues, 37, 3)

Force Field Parameterization

Proxide can automatically assign force field parameters (charges, radii, etc.) via the Rust backend:

from proxide import parse_structure, OutputSpec

# Configure parsing options
spec = OutputSpec(
    add_hydrogens=True,             # Add missing geometric hydrogens
    infer_bonds=True,               # Infer connectivity if missing
    parameterize_md=True,           # Compute MD parameters
    force_field="protein.ff14SB.xml" # Use standard AMBER force field
)

# Parse and parameterize in one step
protein = parse_structure("path/to/structure.pdb", spec)

# Access MD parameters
print(protein.charges)  # Partial charges
print(protein.sigma)    # Lennard-Jones sigma

Trajectory Parsing

from proxide import parse_xtc

# Fast Rust-based XTC parsing
traj_data = parse_xtc("path/to/trajectory.xtc")
coords = traj_data["coordinates"]  # (N_frames, N_atoms, 3)

Fast Partial Charges (Espaloma)

Proxide provides access to high-performance Expaloma graph-neural-network inference for charge assignment using native Rust embedded weights:

# Assign charges with the fast Rust backend
proxide charges molecule.sdf --backend rust

⚡ Performance

The migration to a Rust backend has yielded significant performance improvements compared to the pure Python implementation:

Operation Speedup Example Metric
PDB Parsing 25x -
mmCIF Parsing 25x -
Topology Generation 50x -
Force Field Loading 10x -
Espaloma Inference 7.5x 13.6ms (JAX) ➔ 1.8ms (Rust)

🚀 Deployment & CI/CD

Proxide uses Maturin to seamlessly build Python wheels and publish them to PyPI. To publish proxide releases securely without secrets, we utilize PyPI Trusted Publishing (OIDC). Do not configure a PYPI_API_TOKEN GitHub Secret. Instead, create a Trusted Publisher in the PyPI dashboard bound to this repository to allow .github/workflows/publish.yml to authenticate.


⚠️ Migration Notes

If you are migrating from older versions of Proxide:

  1. Biotite Removal: Direct dependency on biotite for parsing has been removed. All parsing is now handled by _proxider.
  2. API Changes:
    • proxide.io.parsing.biotite -> proxide.parse_structure
    • proxide.physics.force_fields -> proxide.load_forcefield
  3. JAX by Default: Most I/O functions now return JAX arrays by default. Use use_jax=False if you specifically need NumPy arrays.

🔧 Development

Running Tests

# Run all tests
uv run pytest

# Run fast smoke tests
uv run pytest -m smoke

Building Wheels Locally

Before pushing a release to GitHub, test the wheel build locally to catch platform-specific issues early:

# Test the build locally (validates HDF5 static compilation and workflow config)
bash scripts/test-wheel-build.sh

# This will:
# 1. Check build prerequisites (cargo, rustc, python3)
# 2. Build the Rust extension with static HDF5 compilation
# 3. Verify the GitHub workflow is correctly configured

Note: The Rust extension uses static HDF5 compilation via the hdf5-metno crate with the static feature enabled. This means HDF5 is compiled from source during the build and included in the binary, eliminating platform-specific library dependencies.

Linting and Typing

# Linting
uv run ruff check src/proxide/ --fix

# Type Checking
uv run ty check

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

proxide-0.1.0a15.tar.gz (7.3 MB view details)

Uploaded Source

Built Distributions

If you're not sure about the file name format, learn more about wheel file names.

proxide-0.1.0a15-cp311-abi3-win_amd64.whl (14.0 MB view details)

Uploaded CPython 3.11+Windows x86-64

proxide-0.1.0a15-cp311-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (13.7 MB view details)

Uploaded CPython 3.11+manylinux: glibc 2.17+ x86-64

proxide-0.1.0a15-cp311-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl (13.4 MB view details)

Uploaded CPython 3.11+manylinux: glibc 2.17+ ARM64

proxide-0.1.0a15-cp311-abi3-macosx_11_0_arm64.whl (12.5 MB view details)

Uploaded CPython 3.11+macOS 11.0+ ARM64

proxide-0.1.0a15-cp311-abi3-macosx_10_12_x86_64.whl (12.9 MB view details)

Uploaded CPython 3.11+macOS 10.12+ x86-64

File details

Details for the file proxide-0.1.0a15.tar.gz.

File metadata

  • Download URL: proxide-0.1.0a15.tar.gz
  • Upload date:
  • Size: 7.3 MB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.14

File hashes

Hashes for proxide-0.1.0a15.tar.gz
Algorithm Hash digest
SHA256 ae35a30c0988acf9391089d3f8838ce616794658cb53fcad39bb62e811cd830d
MD5 d64eaac4aaca7de4fd5b8dde6795783a
BLAKE2b-256 954f34be1bfd2e5ff5e456167ffa1226db758bf4e2945a322b9ccb41d1bc3696

See more details on using hashes here.

Provenance

The following attestation bundles were made for proxide-0.1.0a15.tar.gz:

Publisher: publish.yml on maraxen/proxide

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file proxide-0.1.0a15-cp311-abi3-win_amd64.whl.

File metadata

  • Download URL: proxide-0.1.0a15-cp311-abi3-win_amd64.whl
  • Upload date:
  • Size: 14.0 MB
  • Tags: CPython 3.11+, Windows x86-64
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.14

File hashes

Hashes for proxide-0.1.0a15-cp311-abi3-win_amd64.whl
Algorithm Hash digest
SHA256 2aa5220a135697793668df08d40777b42a8bdba90e2198338d0d5f6e5bd43ace
MD5 4181e6209f2d48007398199afcf72474
BLAKE2b-256 f471f78ea6b074b5b630c865bfe4bb792d3d46ef5bacbb2564e2ac20e5541b9a

See more details on using hashes here.

Provenance

The following attestation bundles were made for proxide-0.1.0a15-cp311-abi3-win_amd64.whl:

Publisher: publish.yml on maraxen/proxide

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file proxide-0.1.0a15-cp311-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.

File metadata

File hashes

Hashes for proxide-0.1.0a15-cp311-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
Algorithm Hash digest
SHA256 bf1cc605b0ff0236b9672126a13b84d6e754a00a151660aa29ff57a2416f69b2
MD5 2b3051d4189c596890fb55cc166f633d
BLAKE2b-256 7f97a5db53eec400c4eaafbf9f7eed8739e066333672d7f7b71c4c154ae8d3d0

See more details on using hashes here.

Provenance

The following attestation bundles were made for proxide-0.1.0a15-cp311-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl:

Publisher: publish.yml on maraxen/proxide

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file proxide-0.1.0a15-cp311-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl.

File metadata

File hashes

Hashes for proxide-0.1.0a15-cp311-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl
Algorithm Hash digest
SHA256 e7616a3505473e1fd00f6d526e3df59edd891bcf77aee4900aeed0635db0e1a1
MD5 d1faa7f53f17cd60211a3cc331d5c1fc
BLAKE2b-256 384f198ee7b1969f8a455f8c2dca98204eecf93327f48aed24d982a9770f2343

See more details on using hashes here.

Provenance

The following attestation bundles were made for proxide-0.1.0a15-cp311-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl:

Publisher: publish.yml on maraxen/proxide

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file proxide-0.1.0a15-cp311-abi3-macosx_11_0_arm64.whl.

File metadata

File hashes

Hashes for proxide-0.1.0a15-cp311-abi3-macosx_11_0_arm64.whl
Algorithm Hash digest
SHA256 6efe7d1a42fc6584684cb228cc3856d31e40cda103c26e43e32da1a4488e1e97
MD5 7955dc109f544e7df1b0933e26ca83a6
BLAKE2b-256 de2b3eb9cd29a99764b525a85f4822e58fcf89f9f9f58e7524b38badedbe4849

See more details on using hashes here.

Provenance

The following attestation bundles were made for proxide-0.1.0a15-cp311-abi3-macosx_11_0_arm64.whl:

Publisher: publish.yml on maraxen/proxide

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file proxide-0.1.0a15-cp311-abi3-macosx_10_12_x86_64.whl.

File metadata

File hashes

Hashes for proxide-0.1.0a15-cp311-abi3-macosx_10_12_x86_64.whl
Algorithm Hash digest
SHA256 374571407e540b20864619c7a2dea1dc9714b186ed15760bd14686457439906a
MD5 c4736bb66ad063a08afef004c33de7fe
BLAKE2b-256 90c0abd904a715e749f9346b32196c1e1feb9bf01ad4e9fd33a6c7b5c3529531

See more details on using hashes here.

Provenance

The following attestation bundles were made for proxide-0.1.0a15-cp311-abi3-macosx_10_12_x86_64.whl:

Publisher: publish.yml on maraxen/proxide

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.
Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page