Ligand based virtual screening using SVM on GPU.

Area of Science:

• Computational Chemistry
• Bioinformatics
• Drug Discovery

Background:

• In silico methods, particularly virtual screening, are crucial for efficient drug discovery.
• Ligand-based virtual screening models target proteins using known ligand data.
• Support Vector Machines (SVMs) are effective for classification in virtual screening but can be computationally intensive.

Purpose of the Study:

• To develop and benchmark a GPU-based tool for accelerating ligand-based virtual screening.
• To address the computational bottleneck of training SVM models on large ligand datasets.

Main Methods:

• Implementation of a data-parallel virtual screening tool, GpuSVMScreen, utilizing SVMs on Graphics Processing Units (GPUs).
• Benchmarking the performance of GpuSVMScreen for high-throughput screening capabilities.

Main Results:

• GpuSVMScreen demonstrates high throughput, significantly reducing screening time.
• The tool is capable of screening billions of molecules efficiently.
• The source code is publicly available for further research and application.

Conclusions:

• GPU acceleration effectively speeds up SVM-based virtual screening.
• GpuSVMScreen offers a powerful solution for large-scale virtual screening in drug discovery.
• The tool's availability promotes advancements in computational drug design.