A hybrid computational strategy to address WGS variant analysis in >5000 samples

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Decreasing sequencing costs necessitate efficient, real-time variant calling for whole genome sequencing (WGS) data.
• Current computational resources and cloud infrastructures face limitations for large-scale joint variant calling.
• Existing strategies often fail to scale or abandon joint calling for massive datasets.

Purpose of the Study:

• To develop a high-throughput framework for cost-effective and real-time variant calling of WGS data.
• To enable scalable joint variant calling and imputation for large cohorts.
• To address the computational challenges of analyzing large-scale genomic datasets.

Main Methods:

• Leveraged a hybrid computing infrastructure combining cloud (AWS), supercomputers (ORNL, Rice), and local HPC.
• Developed a novel binning approach for large-scale joint variant calling and imputation.
• Utilized multiple state-of-the-art variant callers: SNPTools, GATK-HaplotypeCaller, GATK-UnifiedGenotyper, and GotCloud.

Main Results:

• Successfully performed joint calling, imputation, and phasing for over 5300 WGS samples from the CHARGE dataset in under 6 weeks.
• The framework scales to over 10,000 samples, producing single nucleotide variant (SNV) callsets with high sensitivity and specificity.
• Processed 180 TB of BAM files using 5.2 million core hours across heterogeneous platforms with minimal data transfer (6 TB).

Conclusions:

• Ensemble joint calling of SNVs for low-coverage WGS data is achievable in a scalable, cost-effective, and rapid manner.
• Heterogeneous computing platforms can be effectively utilized for large-scale genomic analyses without compromising variant quality.
• The developed framework overcomes limitations of traditional infrastructures for large-scale genomic data processing.