colabfold_receipe.def

Bootstrap: docker
From: nvidia/cuda:11.1.1-cudnn8-runtime-ubuntu20.04

%setup
#section are first executed on the host system outside of the container after the base OS has been installed. . You can reference the container file system with the $SINGULARITY_ROOTFS environment variable in the %setup section


%files
#The %files section allows you to copy files into the container with greater safety than using the %setup section. Its general form is
#Useless for now.

%environment
  #Define environment variable that will be executed at each runtime.
  export TF_FORCE_UNIFIED_MEMORY=1
  export XLA_PYTHON_CLIENT_MEM_FRACTION=4.0
  export OPENMM_CPU_THREADS=32
  export CUDA=11.1
  export PATH=/opt/mmseqs/bin/:$PATH

  #PATH
  export PATH=/usr/local/cuda-11.1/bin:$PATH
  export LD_LIBRARY_PATH=/usr/local/cuda-11.1/lib64:$LD_LIBRARY_PATH
  export PATH=/opt/conda/bin:$PATH
# XDG_CACHE_HOME needs to be overwritten at runtime
  export XDG_CACHE_HOME=/opt/cache
  export COLABFOLDDIR=/opt/conda/lib/python3.7/site-packages/colabfold


%post -c /bin/bash

  #Redefine the variables
  export CUDA=11.1
  export XDG_CACHE_HOME=/opt/cache
  export COLABFOLDDIR=/opt/conda/lib/python3.7/site-packages/colabfold


  #Install necessary packages
  apt-get update
  DEBIAN_FRONTEND=noninteractive apt-get install -y \
        build-essential \
        cmake \
        cuda-command-line-tools-${CUDA/./-} \
        git \
        hmmer \
        kalign \
        tzdata \
        wget \
        curl \
        vim \
        git \
        libfontconfig1-dev \
        libfreetype6-dev \
        libfreetype6
  rm -Rf /var/lib/apt/lists/*

  #Install MMSEQS
  (cd  /opt;
  wget https://mmseqs.com/latest/mmseqs-linux-avx2.tar.gz;
   tar xvfz mmseqs-linux-avx2.tar.gz;
    rm mmseqs-linux-avx2.tar.gz)
  
  export PATH=/opt/mmseqs/bin/:$PATH



  # Install Miniconda package manager.
  wget -q -P /tmp \
    https://repo.anaconda.com/miniconda/Miniconda3-py37_4.9.2-Linux-x86_64.sh \
      && bash /tmp/Miniconda3-py37_4.9.2-Linux-x86_64.sh -b -p /opt/conda \
      && rm /tmp/Miniconda3-py37_4.9.2-Linux-x86_64.sh

  #Add conda in the PATH.
  export PATH=/opt/conda/bin:$PATH


  # Install necessary soft with conda.
  # conda update -qy conda \
  #    && 
  conda install -y -c bioconda -c conda-forge \
        openmm=7.5.1 \
        pdbfixer \
        pip \
        kalign3=3.2.2 \
        hhsuite=3.3.0
        #cudatoolkit==${CUDA} \



  #Install alignment tools.
  #conda install -c conda-forge -c bioconda kalign3=3.2.2 hhsuite=3.3.0 -y


  #Install matplotlib-inline and ipython
  conda install -y -c conda-forge matplotlib-inline ipython

 
  #Install ColabFold and Jaxlib
  #UPDATE ALPHAFOLD 2.3
  pip install -q --no-warn-conflicts "colabfold[alphafold-minus-jax] @ git+https://github.com/sokrypton/ColabFold"
  pip install https://storage.googleapis.com/jax-releases/cuda11/jaxlib-0.3.25+cuda11.cudnn805-cp37-cp37m-manylinux2014_x86_64.whl


  #Patch Colabfold batch.py to remove environment variable for GPU memory. Those variable are setted up externally.
  (cd  /opt/conda/lib/python3.7/site-packages/colabfold/; wget https://raw.githubusercontent.com/tubiana/colabfold_singularity/main/remove_environ_variable.patch; patch -s -p0 < remove_environ_variable.patch; rm remove_environ_variable.patch)

  # #Apply patches and parameters
  # cd $COLABFOLDDIR
  # sed -i -e "s#props_path = \"stereo_chemical_props.txt\"#props_path = \"${COLABFOLDDIR}/stereo_chemical_props.txt\"#" batch.py
  # wget -q https://git.scicore.unibas.ch/schwede/openstructure/-/raw/7102c63615b64735c4941278d92b554ec94415f8/modules/mol/alg/src/stereo_chemical_props.txt
  # wget https://raw.githubusercontent.com/deepmind/alphafold/main/docker/openmm.patch
  # (cd  /opt/conda/lib/python3.7/site-packages; wget https://raw.githubusercontent.com/deepmind/alphafold/main/docker/openmm.patch; patch -s -p0 < openmm.patch; rm openmm.patch)
  # (cd  /opt/conda/lib/python3.7/site-packages/colabfold/; wget https://raw.githubusercontent.com/tubiana/colabfold_singularity/main/matplotlib.patch; patch -s -p0 < matplotlib.patch; rm matplotlib.patch)

%labels
  author thibault Tubiana

%help
  Singularity for running colabfold localy. 
  NOTE :
   - databases like Uniref30 or other are too big to be place in the containers. Therefore your need to download them and place them outside (check https://colabfold.mmseqs.com). Give this folder later to the picture (see usage).
   - alphafold paramerters are downloaded outsite the container as well. You will need to give a specific folder for that as well.
  
  USAGE:
   1. Define location folder of 
      - MMSEQS: bin folder of MMSEQS
      - DATABASES: Root folder that the database (should have "uniref30" inside etc..)
      - FASTA_DIR: folder that will contain the fasta file
      - PARAMS_DIR: fodler that will contain Alphafold Cache File, correspond to XDG_CACHE_HOME
      - MSA_DIR: Folder that will contain the multiple sequence alignment file
      - PRED_DIR: Folder that will contain the predictions
      - IMAGESINGULAIRTY: alias for the singularity container location.

   2. You can setup a shortcut for the singularity container with:
      SINGULARITYCOMAND="singularity exec \
    -B ${MMSEQS}:/programs\
    -B ${DATABASES}:/alpha/database\
    -B ${FASTA_DIR}:/inout/fasta\
    -B ${PARAMS_DIR}:/opt/cache\
    -B ${MSA_DIR}:/inout/msas -B ${PRED_DIR}:/inout/predictions --nv ${IMAGESINGULARITY}"

  3. Run colabfold_batch and colabfold_search with command like :
  CUDA_VISIBLE_DEVICES=0 $SINGULARITYCOMAND colabfold_batch --model-type /inout/msas /inout/predictions