Nearl is an open-source machine learning framework for mining protein dynamics from molecular dynamics trajectories. In current release, featurization centers on 3D voxel-based representation for 3D-CNN-based frameworks.
- Multiple pre-defined true 3D features including 2 dynamic features and 14 static features
- Flexible definition of features and trajectory container
- Automated pipeline for featurization
- User-friendly API for the customization of features, trajectory suppliers
- Pre-built 3D-CNN models for training and development
The development and tests are performed on Linux(Ubuntu), and the software is not tested on other operating systems. Since the software is still under development and not yet uploaded to PyPI, the installation can be done via direct installation from GitHub repository.
micromamba create -n nearl_env python=3.9.17 AmberTools=23 openbabel=3.1.1
micromamba activate nearl_env
pip install git+https://github.com/miemiemmmm/Nearl
pip install git+https://github.com/miemiemmmm/SiESTA.gitNote
To correctly compile the GPU code, the older device have to adjust the CUDA_COMPUTE_CAPABILITY accordingly, to match the CUDA architecture. The current default value is sm_80.
Install from source:
git clone https://github.com/miemiemmmm/Nearl
cd Nearl
pip install .Runing the following command to check the installation of major components from Nearl:
python -m nearl.valid_installationimport nearl
import nearl.data
loader = nearl.TrajectoryLoader([nearl.data.MINI_TRAJ])
feat = nearl.Featurizer({"dimensions": 32, "lengths":16, "time_window":10})
feat.register_feature(nearl.features.Mass(outkey='mass', outfile="/tmp/test.h5", sigma=1.5, cutoff=5.0))
feat.register_focus([":ARG"], "mask")
feat.register_trajloader(loader)
feat.main_loop()You can find detailed documentation at either of the following locations:
This project is licensed under the MIT License - see the LICENSE file for details.