minifold - GitLite

KO | EN

gitlite — search

#javascript #python #react #go #golang #deep-learning #linux #docker #java #kubernetes #security #ios #mcp #openai

minifold

(★ 94)

MiniFold: Simple, Fast and Accurate Protein Structure Prediction

File Explorer

Download Latest Version (.zip)

- stage1.yaml
- stage2.yaml
- - __init__.py
  - config.py
  - data_pipeline.py
  - data_transforms.py
  - feature_pipeline.py
  - input_pipeline.py
  - mmcif_parsing.py
  - of_data.py
  - parsers.py
- - __init__.py
  - heads.py
  - miniformer.py
  - model.py
  - structure.py
- - __init__.py
  - data.py
  - loss.py
  - model.py
- - esm.py
  - feats.py
  - init.py
  - metrics.py
  - openfold.py
  - profile.py
  - protein.py
  - residue_constants.py
  - rigid_utils.py
  - stereo_chemical_props.txt
  - tensor_utils.py
- __init__.py
.gitignore
example.fasta
LICENSE
predict.py
pyproject.toml
README.md
train.py

# Installation Guide

1. Get the code

git clone https://github.com/jwohlwend/minifold

Downloads the entire project code from GitHub to your computer.

cd minifold

Moves into the project folder you just downloaded.

2. Python

Easy Recommended

Prerequisites

Git Needed to download the project code from GitHub.
Python 3 On Windows, be sure to check 'Add Python to PATH' during installation.

pip install .

Installs the package published on PyPI directly — no need to clone the source.

python <실행할 파일명>.py # README에서 정확한 실행 파일명을 확인하세요

Runs the Python script (or module).

✅ If it runs without errors and prints output in the terminal, it worked.

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# MiniFold

MiniFold if a fast model for single chain protein structure prediction. Built using the same protein language model as ESMFold, it achieves considerable speedups (up to 10-20x) and memory savings (up to 10x) at little to no cost in performance.

[Paper Link](https://openreview.net/pdf?id=1p9hQTbjgo)

## Installation

Then install minifold, clone the repository and install it with:

`pip install .`

> We recommend installing in a fresh python environment.

## Inference

To run prediction, use the following:

`python predict.py example.fasta --out_dir PATH --cache PATH`

> The fasta header will be used to name the output file, set them wisely!

Options are available:

- `kernels`: uses our custom triton kernels
- `compile`: use torch.compile with dynamic shapes enabled
- `model_size`: one of `12L` or `48L` (default) 
- `token_per_batch`: maximum number of tokens that fit in your GPU, by default 2048

## Training

We train the model using AFDB proteins filtered to > 70 global plddt and selected for diversity using uniref30 as initial list, pre-filtering. You may use the provided train.py script and the YAML configs under `configs`.

## Kernels

We developed two triton kernels for this work. You can find them [here](https://github.com/jwohlwend/minifold/tree/main/minifold/model/kernels).

## Cite

```
@article{
  wohlwend2025minifold,
  title={MiniFold: Simple, Fast, and Accurate Protein Structure Prediction},
  author={Jeremy Wohlwend and Mateo Reveiz and Matt McPartlon and Axel Feldmann and Wengong Jin and Regina Barzilay},
  journal={Transactions on Machine Learning Research},
  issn={2835-8856},
  year={2025},
  url={https://openreview.net/forum?id=1p9hQTbjgo},
  note={Featured Certification}
}
```