Time-domain event detection using single-instruction, multiple-thread gpGPU architectures in single-molecule...

Area of Science:

• Computational Biology
• Data Science
• Biophysics

Background:

• Time-domain data with discrete amplitude levels often indicate system's latent states.
• Analyzing ordered data serially limits computational efficiency and hardware acceleration.
• Single-molecule measurements (nanopores, smFETs) generate large, time-ordered datasets.

Purpose of the Study:

• To develop advanced algorithms for efficient analysis of time-domain data.
• To leverage general-purpose graphics processing units (gpGPUs) for accelerated feature detection.
• To overcome limitations of serial processing in large-scale biological datasets.

Main Methods:

• Implementation of three novel gpGPU-based algorithms for time-series analysis.
• Application of algorithms to nanopore and single-molecule field-effect transistor (smFET) data.
• Utilizing single-instruction multiple-data (SIMD) architectures for parallel processing.

Main Results:

• Achieved a 250× improvement in the rate of salient feature detection.
• Demonstrated significant acceleration of event detection in terabyte-scale datasets.
• Enabled efficient analysis of high-throughput single-molecule data.

Conclusions:

• gpGPU-based algorithms offer a substantial advancement for analyzing ordered, time-domain data.
• The developed methods overcome serial processing bottlenecks in biological data analysis.
• Freely available code facilitates broader application in scientific research.