Bacterial Annotation by Learned Representation Of Genes (C++ version)

Overview

Balrog is a prokaryotic gene finder based on a Temporal Convolutional Network. We took a data-driven approach to prokaryotic gene finding, relying on the large and diverse collection of already-sequenced genomes. By training a single, universal model of bacterial genes on protein sequences from many different species, we were able to match the sensitivity of current gene finders while reducing the overall number of gene predictions. Balrog does not need to be refit on any new genome.

Publication

Preprint available on bioRxiv here.

PLOS Computational Biology publication coming soon...

Install Balrog via conda

conda create -n balrog_env python=3.7 -y

conda activate balrog_env
(alternatively: "source activate balrog_env")

conda install balrog mmseqs2 -c conda-forge -c bioconda -y
conda install pytorch torchvision torchaudio cpuonly -c pytorch -y
(alternatively for macOS: "conda install pytorch torchvision torchaudio -c pytorch")

balrog --help

Use Balrog with a GPU (fastest option)

A GPU will accelerate Balrog significantly. Balrog and MMseqs2 can be installed the same as above, but PyTorch will need to be installed according to system-specific instructions here: https://pytorch.org/get-started/locally/

# example pytorch installation with cudatoolkit 11
conda install pytorch torchvision torchaudio cudatoolkit=11 -c pytorch -y

Compile Balrog from source

Building from source allows balrog to use LibTorch (the C++ API of PyTorch), thereby eliminating the requirement for Python and PyTorch.

Install MMseqs2

Balrog depends on MMseqs2 at runtime to help reduce false positive gene predictions. Fortunately, MMseqs2 is well supported on both Linux and MacOS. Detailed installation instructions for MMseqs2 can be found on the MMseqs2 GitHub here

# install by brew
brew install mmseqs2
# install via conda
conda install -c conda-forge -c bioconda mmseqs2
# install docker
docker pull soedinglab/mmseqs2
# static build with AVX2 (fastest)
wget https://mmseqs.com/latest/mmseqs-linux-avx2.tar.gz; tar xvfz mmseqs-linux-avx2.tar.gz; export PATH=$(pwd)/mmseqs/bin/:$PATH
# static build with SSE4.1
wget https://mmseqs.com/latest/mmseqs-linux-sse41.tar.gz; tar xvfz mmseqs-linux-sse41.tar.gz; export PATH=$(pwd)/mmseqs/bin/:$PATH
# static build with SSE2 (slowest, for very old systems)
wget https://mmseqs.com/latest/mmseqs-linux-sse2.tar.gz; tar xvfz mmseqs-linux-sse2.tar.gz; export PATH=$(pwd)/mmseqs/bin/:$PATH

Build Balrog

# Linux
git clone https://github.com/salzberg-lab/BalrogCPP
cd BalrogCPP
git fetch origin
git checkout -b libtorch_local origin/libtorch_local
mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=${PREFIX} ..
make
export PATH=$(pwd):$PATH

# MacOS
git clone https://github.com/salzberg-lab/BalrogCPP
cd BalrogCPP
git fetch origin
git checkout -b libtorch_local origin/libtorch_local
mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=${PREFIX} ..
make
export PATH=$(pwd):$PATH

Balrog also requires zlib, which is likely already installed on your system.

# Most Linux systems 
sudo apt install zlib

# Ubuntu
sudo apt-get install zlib1g
sudo apt-get install zlib1g-dev

# MacOS
brew install zlib