Vina-GPU 2.1

Vina-GPU 2.1 further improves the virtual screening runtime and accuracy with the noval RILC-BFGS and GCS mthods based on Vina-GPU 2.0. Vina-GPU 2.1 includes AutoDock-Vina-GPU 2.1, QuickVina 2-GPU 2.1 and QuickVina-W-GPU 2.1.

Virtual Screening Results

• Runtime comparison of Vina-GPU 2.1 on Drugbank library (partial)

• Accuracy comparison of Vina-GPU 2.1 on Drugbank library (partial)

Compiling and Running

Windows

Build from source file

Visual Studio 2019 is recommended for build Vina-GPU 2.1 from source

1. install boost library (current version is 1.77.0)

2. install CUDA Toolkit (current version: v12.2) if you are using NVIDIA GPU cards

   Note: the OpenCL library can be found in CUDA installation path for NVIDIA or in the driver installation path for AMD

3. add $(ONE_OF_VINA_GPU_2_1_METHODS)/lib $(ONE_OF_VINA_GPU_2_1_METHODS)/OpenCL/inc $(YOUR_BOOST_LIBRARY_PATH) $(YOUR_BOOST_LIBRARY_PATH)/boost $(YOUR_CUDA_TOOLKIT_LIBRARY_PATH)/CUDA/v12.2/include in the include directories

4. add $(YOUR_BOOST_LIBRARY_PATH)/stage/lib $(YOUR_CUDA_TOOLKIT_PATH)/CUDA/lib/x64in the addtional library

5. add OpenCL.lib in the additional dependencies

6. add --config=$(ONE_OF_VINA_GPU_2_1_METHODS)/input_file_example/2bm2_config.txt in the command arguments

7. add NVIDIA_PLATFORM OPENCL_3_0 WINDOWS in the preprocessor definitions if necessary

8. if you want to compile the binary kernel file on the fly, add BUILD_KERNEL_FROM_SOURCE in the preprocessor definitions

9. build & run Note: ensure the line ending are CLRF

Linux

Note: At least 8M stack size is needed. To change the stack size, use ulimit -s 8192.

1. install boost library (current version is 1.77.0)

2. install CUDA Toolkit (current version: v12.2) if you are using NVIDIA GPU cards

   Note: OpenCL library can be usually in /usr/local/cuda (for NVIDIA GPU cards)

3. cd into one of the three methods of Vina-GPU 2.1 ($(ONE_OF_VINA_GPU_2_1_METHODS))

4. change the BOOST_LIB_PATH and OPENCL_LIB_PATH accordingly in Makefile

5. set GPU platform GPU_PLATFORM and OpenCL version OPENCL_VERSION in Makefile. some options are given below:

   Note: -DOPENCL_3_0 is highly recommended in Linux, please avoid using -OPENCL_1_2 in the Makefile setting. To check the OpenCL version on a given platform, use clinfo.

   Macros	Options	Descriptions
   GPU_PLATFORM	-DNVIDIA_PLATFORM / -DAMD_PLATFORM	NVIDIA / AMD GPU platform
   OPENCL_VERSION	-DOPENCL_3_0 / -OPENCL_2_0	OpenCL version 2.1 / 2.0

6. type make clean and make source to build $(ONE_OF_VINA_GPU_2_1_METHODS) that compile the kernel files on the fly (this would take some time at the first use)

7. after a successful compiling, $(ONE_OF_VINA_GPU_2_1_METHODS) can be seen in the directory

8. change --opencl_binary_path in the ./input_file_example/2bm2_config.txt accordingly and type $(ONE_OF_VINA_GPU_2_1_METHODS) --config ./input_file_example/2bm2_config.txt to run one of the Vina-GPU 2.1 method

9. once you successfully run $(ONE_OF_VINA_GPU_2_1_METHODS), its runtime can be further reduced by typing make clean and make to build it without compiling kernel files (but make sure the Kernel1_Opt.bin file and Kernel2_Opt.bin file is located in the dir specified by --opencl_binary_path)

10. other compile options:

Options	Description
-g	debug
-DTIME_ANALYSIS	output runtime analysis in gpu_runtime.log
-DDISPLAY_ADDITION_INFO	print addition information

Enlarge the docking box

The docking box now can be enlarged by

1. Change -DSMALL_BOX into -DLARGE_BOX in Makefile
2. Type make source and $(ONE_OF_VINA_GPU_2_1_METHODS) --config ./input_file_example/2bm2_config.txt
3. Once the tutorial docking is finished, type make clearn and make
4. Now you can enlarge the docking box --size_x/y/z accordingly (see Limitation below)

Structure Optimization

Optimization	Methods	Reference	Documentation
Receptor Preparation	cross-docking	origin paper	Doc
Binding Pocket Prediction	COACH-D	origin paper	Doc
Ligand Optimization	Gypsum-DL	origin paper	Doc

Usage

Arguments	Description	Default value
--config	the config file (in .txt format) that contains all the following arguments for the convenience of use	no default
--receptor	the recrptor file (in .pdbqt format)	no default
--ligand_directory	this path specifies the directory of all the input ligands(in .pdbqt format)	no default
--output_directory	this path specifies the directory of the output ligands	no default
--lbfgs	--rilc_bfgs 0 turns off the RILC-BFGS, --rilc_bfgs 1 turns on the RILC-BFGS`	--rilc_bfgs 1
--thread	the scale of parallelism	5000 for quickvina2-gpu 2.1, 8000 for others
--search_depth	the number of searching iterations in each docking lane	heuristically determined
--center_x/y/z	the center of searching box in the receptor	no default
--size_x/y/z	the volume of the searching box	no default
--opencl_binary_path	this path specifies the directory of the kernel path	no default

Limitation

Arguments	Description	Limitation
--thread	the scale of parallelism (docking lanes)	preferably less than 10000
--size_x/y/z	the volume of the searching box	less than 100/100/100 for AutoDock-Vina-GPU 2.1 and 70/70/70 for other two variants

Citation

• Tang, Shidi, et al. "Vina-GPU 2.1: towards further optimizing docking speed and precision of AutoDock Vina and its derivatives." IEEE/ACM Transactions on Computational Biology and Bioinformatics (2024).
• Ding, Ji, et al. "Vina-GPU 2.0: further accelerating AutoDock Vina and its derivatives with graphics processing units." Journal of chemical information and modeling 63.7 (2023): 1982-1998.
• Tang, Shidi et al. “Accelerating AutoDock Vina with GPUs.” Molecules (Basel, Switzerland) vol. 27,9 3041. 9 May. 2022, doi:10.3390/molecules27093041
• Trott, Oleg, and Arthur J. Olson. "AutoDock Vina: improving the speed and accuracy of docking with a new scoring function, efficient optimization, and multithreading." Journal of computational chemistry 31.2 (2010): 455-461.
• Hassan, N. M. , et al. "Protein-Ligand Blind Docking Using QuickVina-W With Inter-Process Spatio-Temporal Integration." Scientific Reports 7.1(2017):15451.
• Amr Alhossary, Stephanus Daniel Handoko, Yuguang Mu, and Chee-Keong Kwoh. "Fast, accurate, and reliable molecular docking with QuickVina 2. " Bioinformatics (2015): 2214–2216.