mdzip 0.1.1

Compress MD trajectories using deep convolutional autoencoder

MDZip

This Git repository contains MDZip, a program that leverages the power of autoencoders with residual connections to compress molecular dynamics trajectories while reconstructing global and local structural properties with minimal information loss.

---

Author: Namindu De Silva

[Add Description Here]

Dependencies

• wheel
• mdtraj
• torch
• torchvision
• torchaudio
• pytorch-lightning
• scikit-learn
• numpy
• tqdm

Installation

Linux/Windows with CUDA

Create conda environment

conda create -n <my-env> python=3.10
conda activate <my-env>

Install dependencies (recomended for CUDA build)

conda install -c conda-forge mdtraj
pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu117
pip install lightning

Install MolZip

git clone https://github.com/PDNALab/MDZip.git
cd MolZip
python setup.py sdist bdist_wheel
cd dist
pip insall molzip-0.1.0-py3-none-any.whl

Linux/Windows/OSX without CUDA

conda create -n <my-env> python=3.10
conda activate <my-env>
conda install -c confa-forge mdtraj

git clone https://github.com/nami-rangana/MolZip.git
cd MolZip
python setup.py sdist bdist_wheel
cd dist
pip insall molzip-0.1.0-py3-none-any.whl

Help

compress(traj: str, top: str, stride: int = 1, out: str = '/blue/alberto.perezant/t.desilva/MolZip/testing', fname: str = '', epochs: int = 100, batchSize: int = 128, lat: int = 20, w: float = 1.0, memmap: bool = False)

compressing trajectory
----------------------
traj (str) : Path to the trajectory file
top (str) : Path to the topology file
stride (int) : Read every strid-th frame [Default=1]
out (str) : Path to save compressed files [Default=current directory]
fname (str) : Prefix for all generated files [Default=None]
epochs (int) : Number of epochs to train AE model [Default=100]
batchSize (int) : Batch size to train AE model [Default=128]
lat (int) : Latent vector length [Default=20]
w (float) : Non-negative weight for loss function [Default=1.0]
memmap (bool) : Use memory-map to read trajectory [Default=False]

decompress(top: str, model: str, compressed: str, out: str)

decompress compressed-trajectory
--------------------------------
top (str) : Path to the topology file (parm7|pdb)
model (str) : Path to the saved model file
compressed (str) : Path to the compressed trajectory file
out (str) : Output trajectory file path with name. Use extention to define file type (*.nc|*.xtc)

Cite

doi : https://doi.org/10.1101/2025.07.31.667955