F2Dock

Fast Fourier protein–protein docking with multi-term scoring.

F2Dock predicts how two proteins bind by exhaustively sampling rigid-body orientations on an FFT grid and ranking poses with a combined electrostatic, van der Waals, shape complementarity, and solvation score. A companion tool, GB-Rerank, rescores the top poses using a Generalized Born solvation model for improved accuracy.

Developed at the Computational Visualization Center, Oden Institute for Computational Engineering and Sciences, The University of Texas at Austin.

Features

• FFT-accelerated exhaustive rotational/translational search
• Multi-term scoring: electrostatics, van der Waals, desolvation, shape complementarity
• Generalized Born reranking (GB-Rerank)
• Hydrogen-bond and residue-contact post-filters
• Dynamic Packing Grid (PG) for fast spatial range queries
• Cross-platform: Linux, macOS, Windows

Quick start

Prerequisites

Dependency	Notes
C++20 compiler	GCC 13+, Clang 16+, or MSVC 19.4+
cvcpkg	curl -fsSL https://cvcpkg.org/install.sh | sh (Linux, macOS) or irm https://cvcpkg.org/install.ps1 | iex (Windows). It is not on PyPI.

Everything else — libcvc, PocketFFT, GoogleTest, Boost, CGAL, HDF5, ImageMagick, pthreads4w on Windows, and even CMake and Ninja — is installed from the cvcpkg catalog. F2Dock does not use apt, Homebrew or vcpkg, and does not build libcvc from source.

Build

cvcpkg install-deps cvcpkg/recipes/f2dock --prefix deps \
    --config release --link shared --include-host-tools
export PATH="$PWD/deps/bin:$PATH"
cmake -B build -G Ninja -DCMAKE_BUILD_TYPE=Release -DCMAKE_PREFIX_PATH="$PWD/deps"
cmake --build build --parallel

Executables are placed in build/bin/. The dependency set is whatever cvcpkg/recipes/f2dock/recipe.yaml declares, resolved transitively — the same recipe CI uses, and the same one that builds and publishes the f2dock package.

Packaging

cvcpkg/recipes/f2dock builds the f2dock cvcpkg package (PocketFFT, GPL components off, so LGPL-2.1). It is not published to the catalog yet; once it is:

cvcpkg install f2dock --prefix /some/prefix --config release --link shared

FFT provider

FFTW is GPL-2.0-or-later, so a binary that links it is a combined work under the GPL, not F2Dock's LGPL-2.1. The transform is therefore selectable, and the default is not FFTW:

-DF2DOCK_FFT_PROVIDER=	Licence	Notes
pocketfft (default)	BSD-3-Clause	Header-only, from the cvcpkg pocketfft bundle
fftw	GPL-2.0-or-later	Opt-in; configure warns, and the binary reports itself as GPL at startup. Add FFTW to the prefix first: cvcpkg install fftw3 --prefix deps --config release --link shared

Intel oneMKL and NVIDIA cuFFT (proprietary but redistributable, which the LGPL permits linking) are planned; selecting either today is a configure error.

PocketFFT agrees with FFTW to float64 round-off. On the golden integration cases the poses, ranks and RMSDs are identical and scores differ by at most one unit in the last printed digit; see THIRD-PARTY.md. It is slower, by 1.4-1.7x on transform throughput.

Optional GPL components

Two vendored modules are GPL-2.0-or-later while F2Dock itself is LGPL-2.1-only, so they are not built by default:

Module	Feature it backs	Upstream
sparsefft3	useSparseFFT — sparse 3-D FFT	MIT 2000, modified at CVC Lab
libicp	icpRefine — point-to-plane pose refinement	Andreas Geiger, KIT

The default build uses the provider's dense transform instead of the sparse one. It is slower, and it produces the same scores. icpRefine is unavailable. Both parameters are rejected with an explanatory message rather than silently ignored, and the binary prints what it contains at startup:

GPL components: sparseFFT=no libicp=no (build licence: LGPL-2.1-only)
FFT path: dense, provider PocketFFT (BSD-3-Clause)

To build them:

cvcpkg install fftw3 --prefix deps --config release --link shared
cmake -B build -DCMAKE_BUILD_TYPE=Release -DCMAKE_PREFIX_PATH="$PWD/deps" \
    -DF2DOCK_ENABLE_GPL_COMPONENTS=ON -DF2DOCK_FFT_PROVIDER=fftw

sparsefft3 calls FFTW directly, so it needs the FFTW provider. With -DF2DOCK_ENABLE_GPL_COMPONENTS=ON alone, only libicp is built.

The resulting binary is a combined work covered by the GNU GPL v2 or later, not by F2Dock's LGPL-2.1. See THIRD-PARTY.md.

Run

F2Dock parameterFile
F2Dock -score parameterFile
GB-Rerank parameterFile

Usage example

F2Dock reads a plain-text parameter file that lists the input molecules and scoring options. Four example parameter files and their associated test data are included in tests/.

Run docking on the 1ACB enzyme–inhibitor complex from the repository root:

build/bin/F2Dock tests/1ACB.inp

The parameter file tests/1ACB.inp looks like:

staticMolecule   tests/1ACB_R_U_1.7.f2d
movingMolecule   tests/1ACB_L_U.f2d
staticMoleculePQR  tests/1ACB_R_U.pqr
movingMoleculePQR  tests/1ACB_L_U.pqr
staticMoleculeQUAD tests/1ACB_R_U.quad
movingMoleculeQUAD tests/1ACB_L_U.quad
rmsdAtoms        tests/1ACB_rmsd_backbone_unbound_10.0.txt
outFile          tests/1ACB_out.txt

rotFile          deg15.mtx
numRot           100000
effGridFile      fftw-rank.txt

complexType      E
numSolutions     20000
rerank           true
numRerank        2000
breakDownScores  true
numThreads       4

Key parameters

Parameter	Description
staticMolecule / movingMolecule	F2D structure files for receptor and ligand
staticMoleculePQR / movingMoleculePQR	PQR charge/radius files
staticMoleculeQUAD / movingMoleculeQUAD	Surface quadrature point files
rotFile	Euler angle rotation set (e.g. deg15.mtx)
numRot	Number of rotations to sample
effGridFile	Machine-specific FFT grid sizes for best performance
complexType	A (antibody–antigen), E (enzyme–inhibitor), G (other), U (auto-detect)
numSolutions	Max docking poses to output
rerank / numRerank	Enable GB-Rerank rescoring and number of top poses to rescore
outFile	Output file for ranked docking poses
numThreads	Number of CPU threads
useSparseFFT	Sparse FFT instead of the dense transform. Requires -DF2DOCK_ENABLE_GPL_COMPONENTS=ON -DF2DOCK_FFT_PROVIDER=fftw; see Optional GPL components
icpRefine	Point-to-plane ICP refinement of the top poses. Requires -DF2DOCK_ENABLE_GPL_COMPONENTS=ON

All file paths are relative to the current working directory. See doc/params/ for the full parameter reference.

Windows

vcpkg install fftw3:x64-windows pthreads:x64-windows
cmake -B build -DCMAKE_TOOLCHAIN_FILE=%VCPKG_ROOT%/scripts/buildsystems/vcpkg.cmake
cmake --build build --config Release

Testing

cmake -B build -DCMAKE_BUILD_TYPE=Debug
cmake --build build --parallel
cd build && ctest --output-on-failure

89 unit tests cover the spatial index, scoring components, file parsers, and math utilities.

Formatting

F2Dock CI enforces clang-format on changed lines.

Use this before pushing:

git fetch origin
git clang-format <base>

For example, if your PR targets master:

git clang-format origin/master

Install

cmake --install build --prefix /usr/local

Installs executables to bin/ and data files (rotation matrices) to share/F2Dock/data/.

Releases

Pre-built binaries for Linux, macOS, and Windows are published on the Releases page whenever a version tag is pushed.

Project layout

src/
  f2dock/         Main docking driver
  fast-GB/        Fast Generalized Born solvation
  fast-LJ/        Lennard-Jones van der Waals
  fast-hydro/     Hydrophobicity scoring
  fast-PQ/        Skin surface / shape complementarity
  fast-clash/     Steric clash filter
  fast-resCont/   Residue contact filter
  PG-range/       Dynamic Packing Grid spatial index
  fft-utils/      FFT grid construction and correlation
  GB-rerank/      GB-Rerank standalone rescoring tool
  hbondFilter/    Hydrogen-bond post-filter
  math/           Pairing heap, spherical harmonics
  libmol/         Molecule I/O (PDB/PQR/F2D) -- not built; see THIRD-PARTY.md
  utils/          Parameter parsing, timers
inc/              Headers (mirrors src/ structure)
tests/unit/       Google Test suite
doc/              Parameter and dependency documentation

References

• Bajaj, C., Chowdhury, R., and Rasheed, M. A Dynamic Data Structure for Flexible Molecular Maintenance and Informatics. Bioinformatics, 27(1):55–62, 2011.
• Chowdhury, R., Rasheed, M., Keidel, D., Moussalem, M., Olson, A., Sanner, M., and Bajaj, C. Protein-Protein Docking with F2Dock 2.0 and GB-Rerank. PLoS ONE, 8(3):e51397, 2013.

License

GNU Lesser General Public License v2.1 only. The full text is in LICENSE, and it is what CPack ships in generated packages.

Vendored third-party code, its licences, and which of it the default build compiles are inventoried in THIRD-PARTY.md. Two entries there are unresolved and are flagged as such.

Contact

Computational Visualization Center Oden Institute for Computational Engineering and Sciences The University of Texas at Austin