Install

The project runs under Ubuntu 18.04.6, Linux-5.3.0-47-generic-x86_64-with-glibc2.27.

prepare python env

conda env create -n metalnet -f env.yaml

If it doesn't work, try:

conda create -n metalnet python=3.9.16
conda activate metalnet
conda install python-graphviz==0.20.1
conda install pytorch==1.13.1 pytorch-cuda=11.7 -c pytorch -c nvidia
conda install hhsuite==3.3.0 -c conda-forge -c bioconda
pip install autogluon.tabular[all]==0.8.0
pip install biopython==1.81
pip install fair-esm==2.0.0
pip install absl-py
pip install more-itertools

download model weights

1. Download model_1 for metal-binding prediction, model_2 for metal-type prediction.
2. Place model_1 at ./model/train/ and model_2 at ./extra/train/, then unzip them and remove zip files.

Usage

See ./app/example/.

Optional

Optional python packages:

conda install mmseqs2 -c conda-forge -c bioconda
pip install matplotlib_venn==0.11.9
pip install bs4==0.0.1
pip install lxml==4.9.2

Optional executables:

# blast related
wget https://ftp.ncbi.nlm.nih.gov/blast/executables/blast+/LATEST/ncbi-blast-2.14.0+-x64-linux.tar.gz

Patches:

PACKAGE_PATH="/path/to/site-packages"

# add arial related font to matplotlib
cp "./asset/*.ttf" -t $PACKAGE_PATH/matplotlib/mpl-data/fonts/ttf/
rm -r "~/.cache/matplotlib/"

Add PROJECT_DIR = "/path/to/metalnet" in ~/.ipython/profile_default/ipython_config.py as a global variable for use.

We find a bug in MMCIFParser, and it will result in a loss of about 100 chains in dataset. But in this project we use the old version of MMCIFParser, still.