PyGBe: Python and GPU Boundary-integral solver for electrostatics

PyGBe—pronounced pigbē—is a Python library that applies the boundary integral method for biomolecular electrostatics and nanoparticle plasmonics.

PyGBe achieves both algorithmic and hardware acceleration. The solution algorithm uses a Barnes-Hut treecode to accelerate each iteration of a GMRES solver to O(N logN), for N unknowns. It exploits NVIDIA GPU hardware on the most computationally intensive parts of the code using CUDA kernels in the treecode, interfacing with PyCUDA. Some parts of the code are written in C++, wrapped using Cython.

Biomolecular electrostatics:

In this application, PyGBe uses continuum electrostatics to compute the solvation energy for proteins modeled with any number of dielectric regions. The mathematical formulation follows Yoon and Lenhoff (1990) for solving the Poisson-Boltzmann equation of the implicit-solvent model in integral form.

Localized Surface Plasmon Resonance:

PyGBe also uses electrostatics to compute the extinction cross section of scatterers that are much smaller than the incident wavelength. This is relevant, for example, to model localized surface plasmon resonance of nanoparticles, where the quasi-static approximation is valid (Mayergoyz, I. D. and Zhang, Z. (2007), Jung, J., et al (2010)).

Documentation

Detailed documentation is available at http://pygbe.github.io/pygbe/docs/

Installation

Regular installation

The following instructions assume that the operating system is Ubuntu. Run the corresponding commands in your flavor of Linux to install.

Dependencies (last tested)

Python 3.4+ (3.6.1)
Numpy 1.11.1+ (1.13.1)
SciPy 0.17.1+ (0.19.1)
Cython 0.24.1+ (0.24.1)
NVCC 8.0
- gcc 5.4.0
PyCUDA 2017.1.1
matplotlib 1.5.1+ (2.0.2) (optional, for post-processing only)

Python and Numpy

To install the specific version of these packages we recommend using either conda or pip.

To create a new environment for using PyGBe with conda you can do the following:

conda create -n pygbe python=3.6 numpy scipy cython matplotlib
source activate pygbe

and then proceed with the rest of the installation instructions (although note that if you do this, cython is already installed.

Cython

To install Cython we recommend using either conda, your distribution package manager or Cython's website.

NVCC

Download and install the CUDA Toolkit.

PyCUDA

PyCUDA must be installed from source. Follow the instructions on the PyCUDA website. We summarize the commands to install PyCUDA on Ubuntu here:

> cd $HOME
> mkdir src
> cd src
> wget https://github.com/inducer/pycuda/archive/v2016.1.2.tar.gz
> tar -xvzf pycuda-2016.1.2.tar.gz
> cd pycuda-2016.1.2
> python configure.py --cuda-root=/usr/local/cuda
> make
> sudo make install

If you are not installing PyCUDA systemwide, do not use sudo to install and simply run

> make install

as the final command.

Test the installation by running the following:

> cd test
> python test_driver.py

Installing PyGBe

Create a clone of the repository on your machine:

> cd $HOME/src
> git clone https://github.com/barbagroup/pygbe.git
> cd pygbe
> python setup.py install clean

If you are installing PyGBe systemwide (if you installed PyCUDA systemwide), then use sudo on the install command

> sudo python setup.py install clean

PyGBe has been run and tested on Ubuntu 12.04, 13.10, 15.04 and 16.04.

Installation using Docker

Requirements:

Install nvidia-docker, (instructions in their README)
- Check pre-requisites
Follow instructions at the top of Dockerfile.

Run PyGBe

PyGBe cases are divided up into individual folders. We have included a few example problems in examples.

Test the PyGBe installation by running the Lysozyme (lys) example in the folder examples. The structure of the folder is as follows:

lys
  ˫ lys.param
  ˫ lys.config
  ˫ built_parse.pqr
  ˫ geometry/Lys1.face
  ˫ geometry/Lys1.vert
  ˫ output/

To run this case, you can use

> pygbe examples/lys

To test PyGBe-LSPR, run the single silver sphere (lspr_silver) example.

To run lspr cases, you can use

> pygbe-lspr examples/lspr_silver

PyGBe also supports the Solvation-Layer Interface Condition (SLIC) model. Check out the param file in the 1bpi_slic example to see how to modify the different SLIC parameters. Tu run, do

> pygbe-slic examples/1bpi_slic

To run any PyGBe case, you can pass pygbe (or pygbe-lspr if it's a LSPR application or pygbe-slic for SLIC) a relative or an absolute path to the problem folder.

Note that PyGBe will grab the first param and config files that it finds in the problem folder (they don't have to share a name with the folder, but it's helpful for organization). If you want to explicitly pass in a different/specific param or config file, you can use the -p and -c flags, respectively.

If you have a centralized geometry folder, or want to reuse existing files without copying them, you can also pass the -g flag to pygbe to point to the custom location. Note that this path should point to a folder which contains a folder called geometry, not to the geometry folder itself.

For more information on PyGBe's command line interface, run

> pygbe -h

> pygbe-lspr -h

Mesh

In the examples folder, we provide meshes and .pqr files for a few example problems. To plug in your own protein data, download the corresponding .pdb file from the Protein Data Bank, then get its .pqr file using any PDB to PQR converter (there are online tools available for this). Our code interfaces with meshes generated using MSMS (Michel Sanner's Molecular Surface code).

The meshes for the LSPR examples and some Poisson Boltzmann that involve spheres were generated with a script called mesh_sphere.py located in pygbe/preprocessing_tools/.

In Generate meshes and pqr you can find detailed instructions to generate the pqr and meshes.

Performance:

PyGBe Performance

Requirements (latest version tested):

pip install clint (0.5.1)
conda install requests (2.14.2)

References

Barnes, J. and Hut, P. (1986), "A hierarchical O(N log N) force-calculation algorithm," Nature, 324: 446–449, doi: 10.1038/324446a0
Yoon, B.J. and Lenhoff, A.M. (1990), "A boundary element method for molecular electrostatics with electrolyte effects," Journal of Computational Chemistry, 11(9): 1080–1086, doi: 10.1002/jcc.540110911.
Mayergoyz, I. D. and Zhang, Z. (2007). "The computation of extinction cross sections of resonant metallic nanoparticles subject to optical radiation", IEEE Trans. Magn., 43(4):1681–1684,doi: 10.1109/TMAG.2007.892500.
Jung, J., Pedersen, T. G., Sondergaard, T., Pedersen, K., Larsen, A. N., and Nielsen, B. B. (2010), "Electrostatic plasmon resonances of metal nanospheres in layered geometries", Phys. Rev. B, 81(12), doi:10.1103/PhysRevB.81.125413.

Papers published using PyGBe

Cooper, C.D, Bardhan, J.P. and Barba, L.A. (2014), "A biomolecular electrostatics solver using Python, GPUs and boundary elements that can handle solvent-filled cavities and Stern layers," Computer Physics Communications, 185(3): 720–729, doi: 10.1016/j.cpc.2013.10.028, arxiv:1309.4018
Cooper, C.D and Barba, L.A. (2016), "Poisson–Boltzmann model for protein–surface electrostatic interactions and grid-convergence study using the PyGBe code," Computer Physics Communications, 202: 23–32, doi: 10.1016/j.cpc.2015.12.019, arXiv:1506.03745
Cooper, C.D, Clementi, N.C. and Barba, L.A. (2015), "Probing protein orientation near charged nanosurfaces for simulation-assisted biosensor design," Journal of Chemical Physics, 143: 124709 doi: 10.1063/1.4931113, arXiv:1503.08150v4.

Other software

Poisson-Boltzmann Solvers

A few other open-source packages exist for solving implicit-solvent models of the Poisson-Boltzmann equation.

Volumetric-based solvers

Delphi
APBS

Pygbe

Install / Use

README