MELD: Modeling employing limited data

Please cite:

JL MacCallum, A Perez, and KA Dill, Determining protein structures by combining semireliable data with atomistic physical models by Bayesian inference, PNAS, 2015, 112(22), pp. 6985-6990.

Current release info

Name	Downloads	Version	Platforms

Testing

Test versions of MELD are built automatically. Current status:

Installation

MELD can be installed either through conda-forge or from source. Installation through conda-forge is generally simpler and should be preferred.

Conda Forge

First install Miniconda or miniforge by following the appropriate instructions.

If using miniconda, we recommend setting conda-forge as the default channel. (This is already enabled for miniforge.)

conda config --add channels conda-forge 
conda config --set channel_priority strict

We recommend installing MELD into a conda environment. You can name this however you want. We usually name this by the meld version or by the project name, e.g.

conda create -n my-meld-project python
conda activate my-meld-project
conda install meld

This will create and activate an environment called my-meld-project, activate it, and install MELD and its dependencies.

The current supported CUDA versions are 10.2, 11.0, 11.1, and 11.2. By default, conda will install MELD for the higest supported version on your system. On some HPC systems, you may be able to load different versions of the cuda library using the module command or similar. If you need to install MELD for a different version of CUDA than is auto-detected, you can use e.g. conda install cudatoolkit=10.2 meld.

The last step is to install mpi4py, see below.

Building from Scratch

MELD requires a CUDA compatible GPU.

• ambermini or ambertools
• netcdf4
• openmm
• CUDA Toolkit
• python >= 3.10
• numpy
• scipy
• sklearn
• progressbar
• eigen3
• mpi4py (see below)

To install the python portion:

python setup.py install

To install the C++ / CUDA portion:

cd plugin
mkdir build
cd build
ccmake ..
make install
make PythonInstall

Installing mpi4py

MELD requires mpi4py, but does not include it as a dependency, as there are multiple prefered ways to install it, depending on your environment.

If your cluster does not use mpi libraries that are tightly coupled to a high-performance network or to the queuing system, you can simply use the version provided by conda-forge.

To use openmpi:

conda install openmpi mpi4py

To use mpich:

conda install mpich mpi4py

If your cluster uses mpi libraries that are system-specific, you will likely need to compile from source:

pip install --no-deps mpi4py

You may need to load modules and/or configure environment variables for this to work. Consult your system adminstrator or cluster documentation for guidance.

Documentation

There is a limited amount of documentation at meldmd.org. Assistance in building out the documentation is appreciated.