K-mer Analysis of Long-read Alignment Pileups for Structural Variant Genotyping
Structural variant (SV) genotyping, an essential component of genomics, plays a critical role in understanding genetic variation. In recent advancements, the development of kanpig (K-mer ANalysis of PIleups for Genotyping) has introduced significant improvements in this area. Kanpig is a software designed to enhance the accuracy of long-read SV genotyping. Through an intensive benchmarking framework, kanpig’s performance was evaluated using various haplotype-resolved long-read assemblies, resulting in more precise genotypes compared to other long-read SV genotypers.
Kanpig’s algorithm unfolds in four major steps. Initially, it identifies “neighborhoods” by parsing a VCF containing SVs, determining SVs within a defined distance of each other. In the next step, kanpig constructs a variant graph from these neighborhoods with nodes representing SVs and edges connecting downstream, non-overlapping SVs. Subsequently, a BAM file is processed for long-read alignments that cover the neighborhood, generating pileups that are used to identify haplotypes. Finally, a breadth-first search is conducted within the variant graph to find an optimal path that aligns best with identified haplotypes.
A significant innovation of kanpig’s approach lies in its representation of sequences using k-mer vectors, with nodes of the variant graph or a read’s pileups calculated using a small k-value, typically 4 base pairs. This k-mer vector includes counts of all possible k-mers and measures sequence similarity through Canberra distance, allowing for reduced artifacts from sequencing errors. Kanpig’s method was benchmarked for its accuracy and demonstrated a high correlation to traditional sequence similarity measurements, confirming its effectiveness in handling neighboring SVs.
Two crucial components of kanpig’s effectiveness are its clustering of reads into haplotypes before applying them to a variant graph and ensuring variant graphs lack edges connecting overlapping SVs, thus preventing conflicting genotypes. This design ensures consistency and accuracy, exemplified when handling complex SV neighborhoods, such as those involving overlapping deletions, without generating biologically implausible genotypes.
Kanpig’s development benefitted from the evaluation against high-confidence structural variants released by the Genome in a Bottle consortium, namely the GIAB v1.1 benchmarks, to assess its prowess. Assessments revealed that across several experiments, kanpig consistently outperformed other tools in genotyping accuracy, especially with closely positioned or neighboring SVs.
Utilizing data from the Human Pangenome Reference Consortium (HPRC), kanpig was rigorously tested against SVs derived from 47 diverse genome assemblies, validating its accuracy. This ensured that kanpig’s development was optimized using comprehensive and diverse sequencing data, setting a baseline for high-quality SV detection.
The genotyping results showcased that kanpig holds a significant advantage when genotyping SVs particularly in tandem repeats (TRs) or when SVs are closely adjacent. Kanpig demonstrated consistently high performance on samples with various levels of read coverage, remaining more accurate at lower coverages compared to competitors.
Understanding the importance of het/hom (heterozygous/homozygous) ratios in genomic studies, kanpig revealed a balance closest to the true ratio in assemblies. This indicates its superiority in capturing true biological states, as opposed to a stronger bias towards homozygous variants observed in other tools.
Kanpig’s ability to manage errors in variant representations was further explored using tools like sniffles for discovery. Even with potential errors, kanpig excelled in maintaining high genotyping concordance and effectively handled false-positive variants, sustaining precision amidst variances in data quality.
When scrutinized through computational performance tests, kanpig demonstrated an impressive usage efficiency, being both fast and memory-efficient. This makes it an attractive choice for large-scale genomic projects, offering substantial improvements over existing genotypers, which often require more computational resources.
Overall, kanpig shapes as a powerful tool, capable of delivering precise and reliable genotyping across varying genomic data sets. Its innovative use of k-mer vectors for sequence analysis, along with advanced graph-based techniques, marks it as a frontrunner in structural variant genotyping through long-read sequencing.
