Accelerated long-read variant calling with Clair3 for whole-genome sequencing.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Genomic data is rapidly expanding, driving the need for efficient variant calling.
• Long-read sequencing technologies increase computational demands in genomic analysis.
• Deep learning methods offer superior accuracy but are computationally intensive.

Purpose of the Study:

• To develop a computationally efficient framework for accelerated variant calling.
• To improve the speed of deep learning-based variant calling without sacrificing accuracy.
• To support large-cohort genomic studies and time-sensitive clinical applications.

Main Methods:

• Integrated parallelized feature generation, enhanced variant phasing, and in-memory read haplotagging.
• Utilized GPU-accelerated neural network inference for variant calling.
• Dynamically optimized GPU and CPU resource utilization.

Main Results:

• Achieved a 10-20 fold speedup in variant calling for 30× whole-genome sequences.
• Completed variant calling in 12-20 minutes on standard hardware and 12-15 minutes on Apple Mac Studio.
• Maintained state-of-the-art accuracy with SNP F1-scores of 99.32% (ONT) and 99.70% (PacBio).

Conclusions:

• The Clair3 framework provides a rapid, accurate, and scalable solution for variant calling.
• The optimized pipeline addresses the computational challenges of large-scale genomic data analysis.
• This advancement facilitates efficient genomic studies and clinical applications.