Fig. 1: NUMT detection with pangenome graph.

a NUMT detection using short-read sequencing data. b NUMT detection using the pangenome graph. PG-NUMT, Pangenome Graph-based NUMT Detection. c The paired comparison of detected NUMTs absent in the T2T-CHM13 reference genome shows a 2.52-fold increased sensitivity of PG-NUMT compared to the short-read approach (sample size N = 200, two-tailed Welch’s t-test, P = 5.09 × 10⁻⁹⁰). Connecting lines indicate the same samples. d Distribution comparison of length and counts of identified NUMTs between PG-NUMT and the short-read approach. Left, NUMTs uniquely identified by PG-NUMT (n = 123); Right, consensus NUMTs detected by both approaches, excluding two loci uniquely misclassified by short-read (n = 81; Supplementary Fig. 3). From bottom to top: UpSet plot comparing NUMT counts between approaches; violin plot of NUMT length distribution; violin plot of NUMT length distribution (≤ 500 bp). e Cumulative NUMT growth curves depicting sequential assembly addition to the pangenome graph, where colors from light to deep indicate singleton (AC = 1), ultra-rare & rare (AF < 1%), and common (AF ≥ 1%) NUMTs. f Short-read NUMT genotyping across 2,504 unrelated 1KGP individuals by PanGenie within a pangenome graph. g Comparison of allele frequencies between PanGenie genotypes and the pangenome graph across 40 HPRCp1 and 160 APGp1 individuals shows high concordance (Pearson’s r = 0.9998, two-tailed P = 0). The outlier locus (pannumt_109, indicated by an arrow) exhibits reduced genotyping accuracy due to adjacent repetitive sequences and the inversion structure (Supplementary Fig. 12). Box plots denote median and interquartile range (IQR), with whiskers 1.5×IQR. Source data are provided as a Source Data file.