libmolgrid: Graphics Processing Unit Accelerated Molecular Gridding for Deep Learning Applications

Area of Science:

• Computational chemistry
• Machine learning
• Bioinformatics

Background:

• Traditional molecular representations can be limiting for deep learning.
• There is a need for efficient, grid-based molecular data handling.

Purpose of the Study:

• To introduce libmolgrid, a versatile library for 3D molecular data.
• To facilitate machine learning applications in computational chemistry.
• To enable efficient processing of voxelized molecular data.

Main Methods:

• Development of a general-purpose library for voxelized molecular data.
• Implementation of data sampling for machine learning workflows.
• Support for temporal and spatial recurrences for neural networks.
• Optimization for graphics processing units (GPUs).

Main Results:

• libmolgrid provides a novel way to represent 3D molecules.
• The library is designed for seamless integration with deep learning frameworks.
• Optimized performance is achieved through GPU utilization.

Conclusions:

• libmolgrid democratizes grid-based modeling in computational chemistry.
• The library supports advanced neural network architectures.
• It offers an efficient solution for machine learning on molecular data.