Accelerated 2D Classification With ISAC Using GPUs.

Area of Science:

• Cryo-electron microscopy data analysis
• Computational structural biology

Background:

• 2D classification is crucial for analyzing single particles in electron microscopy data.
• Current methods are computationally intensive, especially for large datasets.

Purpose of the Study:

• To present GPU ISAC, a GPU-accelerated version of the Iterative Stable Alignment and Clustering (ISAC) algorithm.
• To enable efficient 2D class averaging on single desktop machines.

Main Methods:

• Developed GPU ISAC, a graphics processing unit (GPU)-accelerated implementation of the ISAC algorithm.
• Utilized consumer-grade GPUs (e.g., Nvidia GeForce GTX 1080 TI) for acceleration.
• Integrated GPU ISAC into the SPHIRE GUI and TranSPHIRE pipeline.

Main Results:

• GPU ISAC achieves performance comparable to twelve high-end cluster nodes using just two consumer-grade GPUs.
• Processes a million particles into class averages within 6-13 hours.
• Demonstrates linear scaling across all input dimensions, ensuring future scalability.

Conclusions:

• GPU ISAC offers a computationally efficient and accessible solution for 2D classification in electron microscopy.
• Accelerates the analysis of large-scale single-particle datasets on standard hardware.
• Provides a user-friendly, open-source tool for structural biology research.