Sequence aligner for accurate template-based protein structure prediction

Template-based modeling, the process of predicting the tertiary structure of a protein by using homologous protein structures, is useful if good templates can be found. Although modern homology detection methods can find remote homologs with high sensitivity, the accuracy of template-based models generated from homology-detection-based alignments is often lower than that from ideal alignments.

We propose a new method that generates pairwise sequence alignments for more accurate template-based modeling. The proposed method trains a machine learning model using the structural alignment of known homologs. When calculating sequence alignments, instead of a fixed substitution matrix, this method dynamically predicts a substitution score from the trained model.

@article{10.1093/bioinformatics/btz483,
    author = {Makigaki, Shuichiro and Ishida, Takashi},
    title = "{Sequence alignment using machine learning for accurate template-based protein structure prediction}",
    journal = {Bioinformatics},
    year = {2019},
    month = {06},
    issn = {1367-4803},
    doi = {10.1093/bioinformatics/btz483},
    url = {https://doi.org/10.1093/bioinformatics/btz483},
}

• Linux (recommended) and macOS X
• Python >=3.6

Open and read example.ipynb.

Install requirements:

pip3 install -r requirements.txt

We also needs FLANN [1] and the Python binding.

git clone https://github.com/mariusmuja/flann
cd flann
### These two procedures are needed only for MacOSX
# brew install cmake hdf5
# git checkout 033b05b1 -b clang
mkdir build
cd build
### PYTHON_EXECUTABLE:FILEPATH and CMAKE_INSTALL_PREFIX
### should follow your environment 
cmake -DCMAKE_INSTALL_PREFIX=~/.local -DPYTHON_EXECUTABLE:FILEPATH=/usr/local/bin/python3 ..
make
make install

For non-root (--user) installation, edit flann/src/python/CMakeLists.txt:

# Around #10
COMMAND ${PYTHON_EXECUTABLE} ${CMAKE_CURRENT_BINARY_DIR}/setup.py install
# Add --user
COMMAND ${PYTHON_EXECUTABLE} ${CMAKE_CURRENT_BINARY_DIR}/setup.py install --user

Data used for kNN index and the paper is availble by dvc pull data.dvc, which requires 140GB disk space.

1. Marius Muja and David G. Lowe, "Fast Approximate Nearest Neighbors with Automatic Algorithm Configuration", in International Conference on Computer Vision Theory and Applications (VISAPP'09), 2009