RaPID2: a parallel scalable framework for identity-by-descent segment detection via parallel PBWT.

Motivation: Identity-by-descent (IBD) detection plays a central role in genetic analysis, supporting applications such as genealogy, population history reconstruction, and disease gene mapping. RaPID, our previously developed IBD detection tool, demonstrated high detection power and accuracy. However, RaPID was not designed with high-performance computing (HPC) in mind and fails to scale efficiently to biobank-scale data. As large-scale datasets become increasingly common, parallel and distributed computing has become essential for maintaining performance at scale. Results: We present RaPID2, a redesigned and scalable IBD detection framework built for both HPC and memory-constrained environments. RaPID2 eliminates the disk I/O bottlenecks of its predecessor, enables memory-aware haplotype pair partitioning, and adopts a parallel and distributed architecture with fault-tolerant execution. It supports both fixed and dynamic window sizes and includes two runtime modes tailored for large-cluster deployment or smaller standalone machines. Our benchmarks show that RaPID2 achieves a 32-fold speedup over the original RaPID at a 2 cM threshold while maintaining statistically identical detection power and accuracy. RaPID2 provides a robust and efficient solution for IBD detection on modern genomic data scales. Availability and implementation: RaPID2 is implemented in C#. Its source code and software package are available at https://github.com/ucfcbb/RaPID2.