Function inference of million-scale microbiomes using multi-GPU acceleration.

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Amplicon sequencing provides microbial community composition but lacks functional insights.
• Current functional inference tools (e.g., PICRUSt2) are computationally intensive and inefficient for large datasets.
• There is a need for high-throughput functional prediction methods to analyze massive microbiome datasets.

Purpose of the Study:

• To introduce MGFunc, a novel GPU-accelerated framework for ultra-high-throughput microbiome functional inference.
• To enable rapid and scalable analysis of microbial functional potential from amplicon sequencing data.
• To overcome computational bottlenecks associated with existing functional prediction tools.

Main Methods:

• MGFunc leverages multi-GPU acceleration for functional inference.
• It transforms amplicon-based functional prediction into standardized matrix multiplication using a genomic content network.
• The framework integrates split data loading, matrix partitioning, and dynamic scheduling for efficient multi-GPU utilization.

Main Results:

• MGFunc achieves speedups of up to several hundred thousand times compared to PICRUSt2.
• It can functionally interpret one million samples within one minute using four GPUs.
• The method demonstrates scalability for analyzing millions of samples with limited GPU memory and RAM.

Conclusions:

• MGFunc provides a highly efficient, low-latency solution for ultra-large microbiome dataset functional inference.
• This framework enables global-scale microbiome studies by facilitating rapid functional analysis.
• The developed software offers a significant advancement for understanding microbial functions in diverse environments and applications.