Fast, parallelized implementation of Principal Component Analysis with constant memory consumption for large data sets.
FastPCA is a PCA-calculator programmed in C++11 parallelized with OpenMP.
The FastPCA package is an implementation of the principal component analysis of large MD data sets, using either Cartesian atom coordinates, interatom distances or backbone dihedral angles as input coordinates. In particular, it features the dihedral angle PCA on a torus (dPCA+) by Sittel et al., 2017, which performs maximal gap shifting to treat periodic data correctly. It is optimized and parallelized with constant memory consumption for large data sets.
For fast matrix diagonalization, LAPACK is used (and needed, of course).
The project includes the ‘xdrfile’ library of GROMACS. Thus, you can use data files written as ASCII data as well as .xtc-trajectories.
For bug-reports just open an issue.
Happy Computing.
The code is published “AS IS” under the simplified BSD license. For details, please see LICENSE.txt
If you use the code for published works, please cite as
This package can be installed with conda via
conda install fastpca -c conda-forge
If conda is not available, it can be compiled as well.
Download this repo
git clone https://github.com/moldyn/FastPCA.git
If gcc = 7.x.x - 8.x.x is used, please use the following branch
git clone --branch fix_gcc8 https://github.com/moldyn/FastPCA.git
For gcc > 9.x.x please use the following branch
git clone --branch fix_gcc10 https://github.com/moldyn/FastPCA.git
Create a build-directory in the project root and change into that directory:
mkdir build
cd build
Run cmake, based on the underlying project:
cmake ..
Hopefully, everything went right. If not, carefully read the error messages. Typical errors are missing dependencies…
If everything is o.k., run make (on multicore machines, use ‘-j’ to parallelize compilation, e.g. ‘make -j 4’ for up to four parallel jobs):
make
Now, you should find the ‘fastca’ binary in the ‘src’ folder.