nPoRe:n-Polymer Realigner for improved pileup variant calling

Abstract

Despite recent improvements in nanopore basecalling accuracy, germline variant calling of small insertions and deletions (INDELs) remains poor. Although precision and recall for single nucleotide polymorphisms (SNPs) now regularly exceeds 99.5%, INDEL recall at relatively high coverages (85×) remains below 80% for standard R9.4.1 flow cells. Current nanopore variant callers work in two stages: an efficient pileup-based method identifies candidates of interest, and then a more expensive full-alignment model provides the final variant calls. Most false negative INDELs are lost during the first (pileup-based) step, particularly in low-complexity repeated regions. We show that read phasing and realignment can recover a significant portion of INDELs lost during this stage. In particular, we extend Needleman-Wunsch affine gap alignment by introducing new gap penalties for more accurately aligning repeated n-polymer sequences such as homopolymers (n = 1) and tandem repeats (2 ≤ n ≤ 6). On our dataset with 60.6× coverage, haplotype phasing improves INDEL recall in all evaluated high confidence regions from 63.76% to 70.66% and then nPoRe realignment improves it further to 73.04%, with no loss of precision.

vcfdist: Accurately benchmarking phased small variant calls in human genomes

“…Figure 2a plots the default parameter configurations for the most commonly used aligners in the 2020 pFDA variant calling challenge. This plot also includes the default parameters for a wider range of tools, such as those used for structural variant (SV) detection (verkko, NGMLR) (21; 22), copy number variant (CNV) detection (nPoRe) (17), edit distance calculation (edlib) (23), and assembly alignment (minimap2’s asm5 and asm10 configurations) (24). The original and normalized affine-gap parameters for each tool configuration are included in Supplementary Figure S6.…”

“…Repetitive sequence compression is common practice for long-read de novo assemblers (21; 25; 26), since incorrect estimation of homopolymer run length is the dominant error mode of nanopore sequencing (27). nPoRe penalizes copy number variation in these repetitive regions only slightly, depending on the copy number and repeat unit length (17). This results in many different points on the left side of Figure 2a.…”

“…As the field shifts towards true whole genome evaluation, we expect variant callsets to frequently include merged results from general-purpose aligners/callers in non-repetitive regions and CNV/SV aligners/callers in more repetitive regions. In low-complexity regions, CNVs and other small gaps are orders of magnitude more likely than elsewhere (17), necessitating the inclusion of variants whose representation falls of the left-hand side of Figure 2a near design point A . As can be seen from Figure 2a, several existing SV and CNV callers already occupy this space.…”

“…As read sequencing technologies have recently improved in terms of both average read length and quality, we are now able to call variants in more difficult regions than ever before (16). INDELs are orders of magnitude more likely to occur in low-complexity repetitive regions (17), and so it is important that variant calling evaluation is not biased against newer tools which allow copy number variation and other common INDELs. Lastly, vcfeval suffers from a few practical constraints such as high memory usage and the inability to evaluate large clusters with many nearby variants due to its exponential complexity (14; 9).…”

Jointly benchmarking small and structural variant calls with vcfdist

“…Figure 14:The design space for affine-gap alignment and variant representation with match, mismatch, gap opening, and gap extension penalties m, x, o, and e. All parameters have been normalized so that m = 0, and the penalties for starting (o + e) and extending (e) a gap are plotted relative to substitutions (x). This plot includes the variant representations used in all four datasets, along with short-read[45], long-read[46,47], assembly[41], edit distance[48], copy number variant[49], and structural variant[50] aligners for comparison. Each aligner is plotted in a unique color, except for when multiple aligners use identical parameters.…”