• AutoDock-GPU is developed by the Forli lab at Scripps Research.
• OpenCL and Cuda accelerated version of AutoDock4.2.6. It leverages its embarrasingly parallelizable LGA by processing ligand-receptor poses in parallel over multiple compute units.
• The OpenCL version was developed in collaboration with TU-Darmstadt and is able to target CPU, GPU, and FPGA architectures. This version itself was based on work done by Imre Pechan from evopro Innovation Kft.
• The Cuda version was developed in collaboration with Nvidia to run AutoDock-GPU on the Oak Ridge National Laboratory's (ORNL) Summit, and it included a batched ligand pipeline developed by Aaron Scheinberg from Jubilee Development.
• A SYCL version, which also supports recent Intel GPUs, is under development as a joint work of TUDa and Intel Link here.

Accelerating AutoDock4 with GPUs and Gradient-Based Local Search, J. Chem. Theory Comput. 2021, 10.1021/acs.jctc.0c01006

See more relevant papers

• Gradient-based local search methods (e.g. ADADELTA), as well as an improved version of Solis-Wets from AutoDock 4.
• Cuda and OpenCL paths to support a wide variety of target platforms based on GPU as well as multicore CPU accelerators.
• Observed speedups of up to 4x (quad-core CPU) and 56x (GPU) over the original serial AutoDock 4.2 (Solis-Wets) on CPU.
• A batched ligand pipeline to run virtual screenings on the same receptor (both OpenCL and Cuda)

Other environments or configurations likely work as well, but are untested.

The first step is to set environmental variables GPU_INCLUDE_PATH and GPU_LIBRARY_PATH, as described here: https://github.com/ccsb-scripps/AutoDock-GPU/wiki/Guideline-for-users

make DEVICE=<TYPE> NUMWI=<NWI>

When DEVICE=GPU is chosen, the Makefile will automatically tests if it can compile Cuda succesfully. To override, use DEVICE=CUDA or DEVICE=OCLGPU. The cpu target is only supported using OpenCL. Furthermore, an OpenMP-enabled overlapped pipeline (for setup and processing) can be compiled with OVERLAP=ON. Hints: The best work-group size depends on the GPU and workload. Try NUMWI=128 or NUMWI=64 for modern cards with the example workloads. On macOS, use NUMWI=1 for CPUs.

After successful compilation, the host binary autodock_<type>_<N>wi is placed under bin.

./bin/autodock_<type>_<N>wi \
--ffile <protein>.maps.fld \
--lfile <ligand>.pdbqt \
--nrun <nruns>

Both options can alternatively be provided in the contents of the files specified with --filelist (-B) (see below for format) and --import_dpf (-I) (AD4 dpf file format).

./bin/autodock_gpu_64wi \
--ffile ./input/1stp/derived/1stp_protein.maps.fld \
--lfile ./input/1stp/derived/1stp_ligand.pdbqt

By default the output log file is written in the current working folder. Examples of output logs can be found under examples/output.

Autostop is ON by default since v1.4. The collective distribution of scores among all LGA populations is tested for convergence every <asfreq> generations, and docking is stopped if the top-scored poses exhibit a small variance. This avoids wasting computation after the best docking solutions have been found. The heuristics set the number of evaluations at a generously large number. They are a function of the number of rotatable bonds. It prevents unreasonably long dockings in cases where autostop fails to detect convergence. In our experience --heuristics 1 and --autostop 1 allow sufficient score evaluations for searching the energy landscape accurately. For molecules with many rotatable bonds (e.g. about 15 or more) it may be advisable to increase --heurmax.

When the heuristics is used and --nev <max evals> is provided as a command line argument it provides the (hard) upper # of evals limit to the value the heuristics suggests. Conversely, --heurmax is the rolling-off type asymptotic limit to the heuristic's # of evals formula and should only be changed with caution. The batch file is a text file containing the parameters to --ffile, --lfile, and --resnam each on an individual line. It is possible to only use one line to specify the Protein grid map file which means it will be used for all ligands. Here is an example:

./receptor1.maps.fld
./ligand1.pdbqt
Ligand 1
./receptor2.maps.fld
./ligand2.pdbqt
Ligand 2
./receptor3.maps.fld
./ligand3.pdbqt
Ligand 3

When the distance-based analysis is used (--contact_analysis 1 or --contact_analysis <R_cutoff>,<H_cutoff>,<V_cutoff>), the ligand poses of a given run (either after a docking run or even when --xml2dlg <xml file(s)> is used) are analyzed in terms of their individual atom distances to the target protein with individual cutoffs for:

• Reactive (default: 2.1 Å): These are interactions between modified atom types numbered 1, 4, or 7 (i.e. between C1 and S4)
• Hydrogen bonds (default: 3.7 Å): Interactions between Hydrogen-bond donor (closest N,O,S to an HD, or HD otherwise) and acceptor atom types (NA,NS,OA,OS,SA atom types).
• Van der Waals (default: 4.0 Å): All other interactions not fulfilling the above criteria.

The contact analysis results for each pose are output in dlg lines starting with ANALYSIS: and/or in <contact_analysis> blocks in xml file output.

Visit the project Wiki.

• If you have a bug report, please check the open issues, and if it has not been reported yet, open a new one.
• If you want to add a new feature, pull/fork the code and submit a pull request.