T2T-CHM13 reference genome reduces mapping bias and enhances alignment accuracy at disease-associated variants Authors: Ilaria Cherchi, Francesco Orlando, Orsetta Quaini, Marta Paoli, Yari Ciani, Francesca Demichelis Published: 2026 ABSTRACT SummaryThe T2T-CHM13v2.0 reference genome added previously uncharacterized genomic sequences and improved the accuracy of repetitive stretches compared to former human genome assemblies. By comprehensive allelic variation analysis and read mapping statistics from sequencing reads aligned to hg38 and T2T-CHM13 assemblies in samples encompassing different sequencing designs, we observed that T2T-CHM13v2.0 assembly significantly reduces the reference mapping bias (RMB) and increases read mapping precision at clinically relevant sites, including BRCA1 pathogenic variants. We report the presence of sequence dissimilarities among reference genomes in the proximity of ClinVar annotated variants, suggesting the need for data re-analysis. Analyzing DNA sequencing data from cancer patients, we identified a mean 2.3% increase in variant allelic frequency (VAF) and 0.3% increase in coverage over clinically relevant variants for samples aligned to T2T-CHM13 compared to hg38. Overall, these findings support the implementation of T2T-CHM13 reference for the improvement of sequencing data analyses in the clinical genomic setting. Introduction The first telomere-to-telomere (T2T) reference genome sequenced from the CHM13 human cell line uncovered approximately 200 million base pairs of genomic sequence, including the short arms of acrocentric chromosomes and centromeric satellite arrays.1 The T2T-CHM13v1.0 assembly corrected falsely duplicated hg38 reference genome regions, leading to improved variant calling across multiple populations.2 Additionally, the v2.0 version provided chromosome Y integration3 and expanded the study of repeat elements,4 epigenetic features,5 and structural variants,6 revealing large-scale differences with previous human genome assemblies.7 Since variations in the nucleotide sequence of a reference genome can considerably affect sequencing read mapping, we sought to specifically assess the effect of aligning reads to T2T-CHM13—instead of to hg38—for performing variant allele frequency (VAF) estimation at single nucleotide polymorphisms (SNPs) and for the analysis of clinically relevant variants, CpG islands, and transcription factor binding sites (TFBS). Further, we reasoned that T2T-CHM13 could significantly ameliorate the bias toward higher mapping rates of reads supporting the reference allele relative to those supporting the alternative allele at heterozygous SNPs, defined as read-mapping bias or reference mapping bias (RMB).8 RMB affects the estimation of allelic frequency at hypervariable sites, e.g., HLA genes,9 and annotated variants, preventing a precise quantification of either inherited or acquired allelic imbalance. Methods developed for mitigating RMB include the alignment of sequencing reads to a global major-allele reference10 or pangenome,11 and the correction of VAF estimates.12 The inflation of reads supporting the reference allele observed in RMB has been attributed to mapping errors arising from the variability among the selected reference genome and the individual’s genomic sequence in the context of ancient DNA analysis.13 We hence hypothesized that aligning human reads to the T2T-CHM13 reference genome would ameliorate the estimation of heterozygous SNPs VAF without requiring RMB correction or additional pre-processing steps. To evaluate the impact of the T2T-CHM13 assembly on the read alignment and on the presence of RMB, we performed in-depth characterization of sequencing data from multiple sequencing designs. We showed that alignment to T2T-CHM13 reference significantly enhances allelic variation estimation and read mapping compared to previous assemblies (e.g., hg38). Mismatches detected in the genomic sequences in the proximity (±150 bp) of SNPs with discordant VAFs across the two assemblies demonstrate that RMB is caused by genomic inaccuracies in the reference genome. We performed an annotation of sequence dissimilarities across reference genomes in UCSC CpG islands, TFBS, and ClinVar variants’ proximity for a prompt evaluation of the necessity for data re-analysis and potential re-design of sequencing probes for clinical applications. We envision that T2T-CHM13 assembly read alignment will facilitate the detection of low VAF variants relevant to precision medicine testing via increased VAF and local coverage at disease-associated genomic sites, possibly improving variant detection. Results Reference mapping bias is significantly reduced in samples aligned to T2T-CHM13 For testing whether aligning reads to the T2T-CHM13 reference genome would reduce RMB and, consequently, affect VAF computation, we aligned a total of 89 samples to hg38 and T2T-CHM13v2.0 assembly and generated pileups on a genome-wide set of 30 million high minor allele frequency (MAF) SNPs (Figure 1A). To provide a comprehensive overview of multiple state-of-the-art sequencing designs, the study included: 25 representative genomic DNA (gDNA) whole-genome sequencing (WGS) data from the 1000 Genomes Project (1kGP; median coverage 30×),14 8 tumor cell line gDNA data from the Cancer Cell Line Encyclopedia (CCLE) WGS15 (median coverage 30×), 26 matched gDNA and cell-free DNA (cfDNA) samples from breast cancer patients sequenced with whole-exome sequencing (WES; median coverage 96× for gDNA and 500× for cfDNA),16 and 30 gDNA and cfDNA samples from prostate cancer patients profiled with a prostate-cancer specific targeted assay (PCF_SELECT12; intended coverage of 200× and 1,000× for gDNA and cfDNA, respectively). Additionally, hg38 and T2T-CHM13 alignment files from N = 1,017 individuals from the recent Oxford Nanopore Technologies (ONT) sequencing of 1kGP17 were analyzed to provide an overview on the impact of reference genome change in long-read sequencing data. First, comparative analysis of read alignment metrics was performed on a sample basis for hg38- and T2T-CHM13 aligned data (Figure S1A). Previously reported differences in mapping statistics for T2T-CHM13 versus hg38 aligned samples2,18 are overall in line with observed trends across all sequencing designs (Figure 1B). A marked increment in read average quality (mean of 5% increase) and a decrease in error rate (mean of 10.7% decrease) was registered (Figure 1B).Figure 1Alignment to T2T-CHM13v2.0 reference genome improves mapping metrics and significantly reduces RMB across diverse sequencing designs(A) Sketch summarizing the cohorts analyzed in this study.(B) Paired mapping statistics for all hg38- and T2T-aligned samples. Each line connecting dots (jitter applied for visualization purposes) corresponds to a sample. Wilcoxon paired test significance is shown for each design/metric pair.(C) Sketch of RMB (left) and computational analysis performed to explore VAF status changes and mapping features of reads spanning SNPs (right).(D) Reference mapping bias analysis for 1kGP VCF entries shared among hg38 and T2T-CHM13 with at least 60 supporting reads, WES and panel-based sequencing samples. The top figure compares peak values for hg38 and T2T with respect to the expected threshold of 0.5, highlighted by a dotted vertical line. Difference between the expected threshold and median peak value among cohorts is reported for both hg38 and T2T-CHM13 alignment. The bottom figure includes the densities of all analyzed heterozygous SNPs VAF values and reports the numeric value of each density peak, corresponding to the dotted vertical lines. The expected VAF value of 0.5 is indicated by the continuous gray vertical line and the value for the RMB peak is depicted by dotted lines colored by assembly.(E) Barplots provide a classification for each heterozygous SNP based on VAF comparison among hg38 and T2T-CHM13. The focus on SNPs with incoherent VAF among assemblies reports the number of entries with VAF increase for hg38 (in dark red) and T2T-CHM13 (orange). One-tailed Wilcoxon test p values for VAF T2T-CHM13 > VAF hg38 are reported on the top of the bars. The annotation on the bottom highlights increasing coverage among the cohorts. For all figures containing boxplots, the center line indicates the median, the box limits represent the first and third quartiles, and whiskers extend to 1.5 times the interquartile range. Alignment to T2T-CHM13v2.0 reference genome improves mapping metrics and significantly reduces RMB across diverse sequencing designs (A) Sketch summarizing the cohorts analyzed in this study. (B) Paired mapping statistics for all hg38- and T2T-aligned samples. Each line connecting dots (jitter applied for visualization purposes) corresponds to a sample. Wilcoxon paired test significance is shown for each design/metric pair. (C) Sketch of RMB (left) and computational analysis performed to explore VAF status changes and mapping features of reads spanning SNPs (right). (D) Reference mapping bias analysis for 1kGP VCF entries shared among hg38 and T2T-CHM13 with at least 60 supporting reads, WES and panel-based sequencing samples. The top figure compares peak values for hg38 and T2T with respect to the expected threshold of 0.5, highlighted by a dotted vertical line. Difference between the expected threshold and median peak value among cohorts is reported for both hg38 and T2T-CHM13 alignment. The bottom figure includes the densities of all analyzed heterozygous SNPs VAF values and reports the numeric value of each density peak, corresponding to the dotted vertical lines. The expected VAF value of 0.5 is indicated by the continuous gray vertical line and the value for the RMB peak is depicted by dotted lines colored by assembly. (E) Barplots provide a classification for each heterozygous SNP based on VAF comparison among hg38 and T2T-CHM13. The focus on SNPs with incoherent VAF among assemblies reports the number of entries with VAF increase for hg38 (in dark red) and T2T-CHM13 (orange). One-tailed Wilcoxon test p values for VAF T2T-CHM13 > VAF hg38 are reported on the top of the bars. The annotation on the bottom highlights increasing coverage among the cohorts. For all figures containing boxplots, the center line indicates the median, the box limits represent the first and third quartiles, and whiskers extend to 1.5 times the interquartile range. The extent of RMB as a function of the reference genome considered was assessed by comparing the VAF distributions of heterozygous SNPs against the expected threshold at 0.5, and annotating SNP features among hg38- and T2T-aligned samples, including VAF status across assemblies and comparison of SNP supporting reads (Figure 1C). While comparing intra-individual VAF standard deviation and coverage over high minor allele frequency (MAF) heterozygous SNPs (i.e., MAF > 0.2 across all major ethnicities and 0.2 ≤ VAF ≤ 0.8; Table S2), larger VAF differences were observed at SNPs with lower coverage values (Figure S2A). In concordance with trends observed both in theoretical and observed VAF distributions at varying coverage thresholds (Figure S2B), samples sequenced at 30× coverage (i.e., WGS 1kGP and CCLE samples) did not exhibit RMB (Figure S2C), albeit still underlying the presence of several SNPs with varying VAF among the two assemblies (18.5%). To expand the study of RMB to higher coverage data from WGS samples, we extracted VAF estimates from the 1kGP VCFs genotype calls (n = 3,202 individuals) for hg3814 and T2T-CHM13v2.0.2 We retained only positions annotated as high MAF SNPs, present in both VCFs, and supported by more than 60 reads (i.e., local coverage > 60). VAF distributions from 1kGP VCFs showed a peak closer to the expected heterozygous value in T2T-CHM13 estimates compared to hg38 (0.497 and 0.492, respectively, bootstrap p value = 2e-04) (Figure 1D). Given that complex genomic regions and segmental duplications (SD), which have been reported to affect ∼7% of the genome,19 can alter coverage and read-mapping, we then excluded from the 1kGP VCF hetSNPs data all the SNPs within SDs (∼10%) and non-syntenic regions between hg38 and T2T-CHM13 (∼0.05%) and we confirm the previously observed trend (Figure S2D). Focusing on high-coverage samples from WES and a custom targeted panel (∼3 Mbp) sequencing, we observed that heterozygous SNPs VAF density curves from samples aligned to T2T-CHM13 were centered around the expected 0.5 value, de facto abolishing RMB. In contrast, hg38-aligned samples exhibited a peak at 0.48, indicating a preference for the reference allele that is concordant with the presence of RMB. SNPs VAF status was classified as incoherent VAF or coherent VAF when the absolute VAF difference (delta VAF, i.e., |VAF T2T - VAF hg38|) among assemblies was greater or was either lower or equal to 0.005, respectively. 54.62% of SNPs from 1kGP VCFs, 21.33% of SNPs in breast cancer WES samples, and 36.94% of SNPs in prostate cancer samples sequenced with the custom targeted panel were classified as sites with incoherent VAF (Figure 1E), underlining the great extent of variation among estimated VAF values from hg38 to T2T-aligned data. A global significant increase in the support toward the alternative allele for T2T-CHM13 was observed at sites with incoherent VAF among assemblies (p value = 1.087e-05 for 1kGP VCF data and p value < 2.22e-16 for WES and custom targeted panel data one-tailed Wilcoxon paired test for VAF T2T-CHM13 > VAF hg38), with a higher number of SNPs with VAF increase in T2T-CHM13 alignment observed in samples with higher coverage values (Figure S2E). RMB shifts and differences in VAF status were also observed when stratifying 1kGP VCF data by super-population groups (Figure S2F). Wilcoxon one-tailed test for testing VAF T2T-CHM13 > VAF hg38 for hetSNPs from 1kGP VCF data is significant for the European (EUR, p value = 1.32e-07), South Asian (SAS, p value = 3.62e-04), and East Asian (EAS, p value = 0.0259) population, and not significant for the American (AMR, p value = 0.648) and African (AFR, p value = 0.648) population. These results are in line with the ancestry of the CHM-13 cell line, which is predominantly of EUR descent with traces of SAS and EAS admixture.1 To pinpoint the cause of the improved mapping on alternative alleles in T2T-CHM13, we analyzed the mapping features of the reads spanning heterozygous SNPs with consistent VAF status, i.e., incoherent VAF or coherent VAF classification in all breast cancer samples (Table S3). Of the hg38 SNPs with consistent VAF status, 5.1% were either mapped to a different chromosome or resulted unmapped, and 18.5% were mapped to another position within the same chromosome when aligned to T2T (Figures S3A and 3B). To determine if VAF shifts are also associated with mapping differences related to reference sequence variation, we extracted the DNA flanking sequences (±150 bp) from each previously characterized SNP with a consistent VAF status. We performed pairwise alignment among T2T-CHM13 and hg38 sequences (STAR Methods). SNPs supported by at least one different sequencing read (discordant support) exhibited a significantly higher proportion of sequence mismatches with respect to SNPs with concordant support (Wilcoxon one-tailed test p value = 2.4e-05; Figure S3D). Overall, these data confirm that the reference genome sequence discrepancies result in differential mapping, with T2T-CHM13 globally reducing the difference between expected versus observed VAFs, thus abolishing the RMB. Comparison of genomic sequences from functionally annotated regions reveals differences between hg38 and T2T-CHM13 Since differences in sequences proximal to SNPs were linked to inconsistencies in mapping, we inquired whether other genomic annotated regions might suffer from mapping biases. We analyzed the UCSC tracks for CpG islands,20 ENCODE ChIP TFBS peaks,21 and the proximal sequence of variants annotated in the ClinVar22 database (±150 bp) through pairwise sequence alignment to annotate regions where sequence discrepancies occur among hg38 and T2T-CHM13 (Figure 2A; Table S4). Observed mismatch rates range from 16.57% for flanking regions of ClinVar variants to 37.64% for CpG islands, while insertions and deletions (i.e., InDels) were present at a similar extent throughout the tracks (min: 1.89%, max: 5.89%; Figure S4A). The same trends were observed when comparing hg19 to the T2T reference genome, while no considerable differences were observed between hg19 and hg38 (∼1% mismatches and <0.5% insertions and deletions) (Figure S4A).Figure 2Comparison of genomic sequences from annotated regions reveals variation among hg38 and T2T-CHM13(A) Summary of pairwise alignment analysis of regulatory and genomic variation elements. Left: sketch presenting the genomic annotation tracks analyzed through pairwise alignment. (Created with BioRender.) Right: barplot reporting the proportion of annotation tracks entries with at least one modification (insertions, deletions, and mismatches) comparing hg38 and T2T-CHM13 sequences.(B) Analysis of ClinVar variants' sequence proximity for hg38vsT2T-CHM13 comparison. Top: sketch of sequence differences within and outside homopolymer stretches. Bottom: ClinVar IDs stratification based on pairwise sequence alignment classification and variant classification. The total percentage of entries with sequence differences is reported on the right of each barplot.(C) Barplot reporting actionable genes included in COSMIC Actionability signature with at least 20% of ClinVar pathogenic variants with sequence differences in their proximity. The total number of variants with differences per gene is reported within each bar and the total percentage of pathogenic variants on the top of the bar. Bars are colored according to the top KEGG pathway hit per gene.(D) Proportion of pathogenic variants by hg38vsT2T-CHM13 pairwise alignment analysis outcome for all MEDGEN disease type present in ClinVar with >10 pathogenic variants and at least one variant with sequence differences.(E) Number of ClinVar IDs intersecting with mutation calls from MSK-IMPACT 50k 2026 and TCGA GDC 2025 cohorts. The percentage of ClinVar IDs for each proximal sequence status comparing hg38 to T2T-CHM13 is reported for each bar. Comparison of genomic sequences from annotated regions reveals variation among hg38 and T2T-CHM13 (A) Summary of pairwise alignment analysis of regulatory and genomic variation elements. Left: sketch presenting the genomic annotation tracks analyzed through pairwise alignment. (Created with BioRender.) Right: barplot reporting the proportion of annotation tracks entries with at least one modification (insertions, deletions, and mismatches) comparing hg38 and T2T-CHM13 sequences. (B) Analysis of ClinVar variants' sequence proximity for hg38vsT2T-CHM13 comparison. Top: sketch of sequence differences within and outside homopolymer stretches. Bottom: ClinVar IDs stratification based on pairwise sequence alignment classification and variant classification. The total percentage of entries with sequence differences is reported on the right of each barplot. (C) Barplot reporting actionable genes included in COSMIC Actionability signature with at least 20% of ClinVar pathogenic variants with sequence differences in their proximity. The total number of variants with differences per gene is reported within each bar and the total percentage of pathogenic variants on the top of the bar. Bars are colored according to the top KEGG pathway hit per gene. (D) Proportion of pathogenic variants by hg38vsT2T-CHM13 pairwise alignment analysis outcome for all MEDGEN disease type present in ClinVar with >10 pathogenic variants and at least one variant with sequence differences. (E) Number of ClinVar IDs intersecting with mutation calls from MSK-IMPACT 50k 2026 and TCGA GDC 2025 cohorts. The percentage of ClinVar IDs for each proximal sequence status comparing hg38 to T2T-CHM13 is reported for each bar. Focusing on CpG islands, we quantified that 40.3% of CpGs have at least one base difference (either a mismatch or an InDel) between hg38 and T2T-CHM13 of which ∼63% show a different number of C bases (Figure S4B). When assessing the potential impact on methylation analyses, we observed that 0.2% of the reads reported a change in methylation levels, either being cytosines or changes in the context (CpG, CHG, CHH), suggesting that differences in the assembly do not significantly impact downstream analyses (Figure S4C; STAR Methods). Negligible effects were observed when performing TF enrichment analysis of TF ChIP-seq peaks’ sequences (Figures S4D and 4E; STAR Methods). To investigate whether discrepancies in the proximal sequence of ClinVar variants were enriched within repetitive regions, we annotated the presence of homopolymeric repeats—defined as repetitive stretches (≥5 bp) of the same base. Cumulatively, 81.8% of ClinVar variants’ proximal sequences (±150 bp) harbor at least one homopolymer stretch, which varies among hg38 and T2T-CHM in 2.6% of variants (Figure S5A), corresponding to 2.1% of all ClinVar variants. The analysis of the proximal DNA sequence of ClinVar annotated variants highlighted the presence of different nucleotide compositions among reference genomes for ∼17% of the variants (Figure 2A), with 15% of ClinVar pathogenic variants with at least one mismatch or indel in their proximity when comparing hg38 and T2T-CHM13 sequences (Figure 2B). In particular, 8% (23 out of 304) of the actionable genes included in the COSMIC Actionability v1423 list harbor more than 20 pathogenic variants with sequence discrepancy among hg38 and T2T-CHM13 in their proximity, suggesting that correct variant detection might depend on the choice of the reference genome. Among genes especially affected by these differences (i.e., >5% of pathogenic variants with proximal sequence modifications and encompassing >100 ClinVar entries), we identified genes involved in cancer pathways, e.g., BRCA1, ATM, MSH2, CHEK2, FANCA, BRCA2, as well as other disease-related pathways (Figures 2C; Figure S5C). We analyzed the MEDGENE code annotation within the ClinVar VCF file to extract disease information and observe the incidence of variation within and outside homopolymers among all pathogenic and likely pathogenic variants per disease (Figure 2D). Twenty-two annotated diseases, including PRPH2-disorder and beta thalassemia, have at least 15% of pathogenic and likely pathogenic variants with differences outside homopolymers. To estimate the abundance of ClinVar mutation calls overlapping sites affected by a sequence change in large cancer cohorts, we intersected SNV calls for MSK-IMPACT 50k 202624 (N patients = 48,179) and TCGA GDC 202525 (N patients = 10,673) pan-cancer datasets with the previously presented ClinVar annotation. Overall, we observed that ∼14.2% of ClinVar IDs demonstrate a sequence change between hg38 and T2T-CHM13 (Figure 2E), including variants that are part of clinical diagnostics, e.g., 43590, 43589, and 127819 ClinVar IDs in TP53 gene for Li-Fraumeni syndrome (Table S7). Reference genome sequence differences within homopolymeric stretches help elucidate differences in SNV VAF estimates among hg38- and T2T-aligned samples To evaluate the effect of the reference genome on the detection of clinically relevant SNVs, we performed variant calling on a set of cfDNA samples from prostate and breast cancer patients using Strelka226 and BCFtools27 (Figure 3A; Table S5). Focusing on ClinVar sites with variations in their proximal sequences, both Strelka2 and BCFtools identified more pathogenic variants in patients’ data aligned to T2T-CHM13 than to hg38 (22.6% and 24.7% increments for Strelka2 and BCFtools, respectively; Figure 3B). Additionally, a significant increase in the quality of SNV calls for T2T-aligned samples (p value = 0.001 and p value < 2.22e-16) was observed with both variant calling methods. The consensus among variant callers over pathogenic and likely pathogenic variants was higher in T2T-aligned samples when compared to hg38-aligned, with 546 (75.8%) and 402 (59.1%) shared calls among variant callers, respectively (Figure S6A).Figure 3Variant calling over ClinVar sites shows improved SNV detection for samples aligned to T2T-CHM13(A) Schematic representation of variant calling analyses performed on cell-free DNA samples from tumor patients (WES and 3 Mb panel) on ClinVar sites.(B) SNV calls with at least 5% VAF and 50 supporting reads performed with Strelka2 and BCFtools. Left: number of identified SNVs stratified by ClinVar classification. Right: quality score for SNV calls performed with Strelka2 and BCFtools. One-tailed pairwise Wilcoxon test for T2T-CHM13 > hg38 p value is reported for both variant calling methods. Outliers not shown. The number of data points is indicated at the bottom of each boxplot.(C) Strelka2 variant calling. Left: boxplot for VAF estimation from variants with and without sequence differences in their 150 bp proximity. Wilcoxon test p value was computed for all comparisons. Right: barplot annotation reporting the percentage of SNVs with differences within their homopolymeric stretch. Homopolymers were defined as stretches of ≥5 equal bases. Fisher's exact test p value is reported.(D) IGV-like representation of differential read support over selected ClinVar variants in BRCA1 (left) and TP53 (right) gene. Top: gene exon structure (hg38 coordinates) and hg38vs T2T-CHM13 BLAST comparison of the entire gene sequence, colored by detected sequence differences. Total length of the analyzed sequence and percentage of dissimilar bases are reported on the top. Bottom: genomic coordinates, coverage profile, reads supporting and not supporting ClinVar variant allele (“Other reads”), and reference sequence composition for hg38 and T2T-CHM13 reference genome. The variant of interest is highlighted in red and VAF estimate is reported for each alignment. For all figures containing boxplots, the center line indicates the median, the box limits represent the first and third quartiles, and whiskers extend to 1.5 times the interquartile range. Variant calling over ClinVar sites shows improved SNV detection for samples aligned to T2T-CHM13 (A) Schematic representation of variant calling analyses performed on cell-free DNA samples from tumor patients (WES and 3 Mb panel) on ClinVar sites. (B) SNV calls with at least 5% VAF and 50 supporting reads performed with Strelka2 and BCFtools. Left: number of identified SNVs stratified by ClinVar classification. Right: quality score for SNV calls performed with Strelka2 and BCFtools. One-tailed pairwise Wilcoxon test for T2T-CHM13 > hg38 p value is reported for both variant calling methods. Outliers not shown. The number of data points is indicated at the bottom of each boxplot. (C) Strelka2 variant calling. Left: boxplot for VAF estimation from variants with and without sequence differences in their 150 bp proximity. Wilcoxon test p value was computed for all comparisons. Right: barplot annotation reporting the percentage of SNVs with differences within their homopolymeric stretch. Homopolymers were defined as stretches of ≥5 equal bases. Fisher's exact test p value is reported. (D) IGV-like representation of differential read support over selected ClinVar variants in BRCA1 (left) and TP53 (right) gene. Top: gene exon structure (hg38 coordinates) and hg38vs T2T-CHM13 BLAST comparison of the entire gene sequence, colored by detected sequence differences. Total length of the analyzed sequence and percentage of dissimilar bases are reported on the top. Bottom: genomic coordinates, coverage profile, reads supporting and not supporting ClinVar variant allele (“Other reads”), and reference sequence composition for hg38 and T2T-CHM13 reference genome. The variant of interest is highlighted in red and VAF estimate is reported for each alignment. For all figures containing boxplots, the center line indicates the median, the box limits represent the first and third quartiles, and whiskers extend to 1.5 times the interquartile range. Importantly, we observed a significant increase in VAF estimates for T2T-aligned samples data when compared to hg38-aligned data at sites with sequence variation (one-tailed Wilcoxon test, p value < 2.22e-16), while entries with no modifications in their proximal sequence showed coherent VAF (Figures 3C; Figure S6B). Analyzed variants not only include somatic SNVs but also inherited germline SNVs and can also coincide with annotated dbSNP sites, suggesting that the VAF improvement observed in samples aligned to T2T-CHM13 partly reflects RMB reduction. To remove potential sample-specific biases, we retained only SNVs with a consistent impact of reference genome change on VAF (i.e., VAF always coherent or incoherent between assemblies across all samples), and we observed a clear enrichment (Fisher test, p value < 2.22e-16) for sequence differences within and outside homopolymeric stretches across sites with discordant VAF (Figure 3C). In-depth characterization of read alignment at ClinVar sites across regions with homopolymer variation highlighted a differential read support for hg38- and T2T-aligned samples, impacting VAF estimates and coverage homogeneity across well-characterized genes, such as BRCA1 and TP53 (Figure 3D). Reads overlapping sites with diverging sequence among assemblies were classified as InDels upon hg38 alignment but not upon T2T-CHM13 alignment, suggesting that the sequence present in T2T-CHM13 assembly better recapitulates the sequence present in the patient’s genome. The BRCA1 gene structure is impacted by numerous changes between hg38 and T2T, resulting in a modified read mapping at both exonic and intronic regions (Figure S6C). A coverage increment was observed at likely pathogenic and pathogenic variants within BRCA1 exon 17 for T2T-aligned samples. Overall, 24% of genes reporting pathogenic or likely pathogenic variants in ClinVar exhibited increments in the VAF of their pathogenic or likely pathogenic variants located on regions of genomic sequence discrepancy (Figure S6D), supporting the notion that adoption of the T2T-CHM13 reference genome improves VAF estimation at numerous genomic sites of clinical interest. To investigate whether the genomic sequence from diverse individuals at these sites is closer to T2T-CHM13 or hg38 reference sequence, we performed a read consensus analysis across 1,017 1kGP long-read individual sample data. Sequence similarity computed between read consensus sequences from the 1kGP long-read cohort and the reference genome over pathogenic or likely pathogenic variants was significantly higher for T2T-aligned samples with respect to hg38-aligned samples (one-tailed Wilcoxon paired test for T2T similarity > hg38 similarity p value < 2.22e-16), further underlining the higher fidelity of T2T-CHM13 to individuals’ genomic sequences against hg38 at these sites irrespectively of super-population groups (Figure S6E). Discussion The choice of the reference genome is crucial for analyzing sequencing data, as it significantly affects read alignment accuracy and the outcome of downstream analyses. The T2T-CHM13 genome build enables significant advances in genetics and genomics studies relevant to human diseases by uncovering previously uncharacterized regions using cutting-edge third-generation sequencing technologies and novel assembly methods. We systematically investigated whether significant differences in genomic feature analysis would be observed when changing the reference genome. Based on the analysis of both in-house and publicly available DNA sequencing datasets across diverse sequencing designs, we underlined the value of aligning human sequencing data to T2T-CHM13v2.0 reference to reduce RMB and globally enhance read mapping. Discrepancies in VAF estimates were consistently present across all analyzed samples and assay designs when mapped on hg38 in respect to T2T-CHM13: globally, aligning to the T2T-CHM13 reference genome improves read mapping and SNP VAF estimation, regardless of the sequencing strategy or intended coverage. We envision that the current efforts to build a comprehensive human T2T pangenome28 will further improve read mapping accuracy and aid in the characterization of the human genome landscape. Furthermore, the analysis of increasingly growing long-read sequencing datasets requires alignment to complete human genomes for the identification of structural variants and complex genetic variations.17,29 Through pairwise sequence comparison analysis, we annotated sequence differences in CpG islands, TFBS peaks, and ClinVar sites, and provide a useful resource for re-evaluating sites likely prone to differential mapping (available on Zenodo and as a ClinVar OpenCRAVAT30 annotation). In this study, while genomic differences within ChIP-seq peaks and CpG islands had no significant impact on TFBS enrichment analyses or methylation calling outcomes, coverage and VAF increment at ClinVar variants suggest that the latest reference genome could, in principle, provide an advantage for variant identification with compared to hg38. Indeed, realigning reads to T2T-CHM13v2.0 assembly might improve the detection of well-characterized pathogenic variants included in genetic tests for therapy selection, particularly for variants with low VAF. As reported in a recent publication on variant detection from Swedish population sequencing data,18 T2T-CHM13 alignment enhances the identification of disease-causing variations and improves read alignment metrics. In our study, by analyzing the sequences proximal to ClinVar annotated variants, we identified the presence of widespread discrepancies among reference genomes as the cause of the previously observed improvements in mapping accuracy at clinically relevant sites. In addition, sequence differences detected in homopolymer stretches and low-complexity regions between hg38 and T2T-CHM13 underscore the importance of reference genome choice for designing DNA probes, which could greatly benefit from a more accurate sequence for improving affinity and sequencing yield. We estimate that ∼14% ClinVar variants currently detected in large pan-cancer cohorts (e.g., MSK-IMPACT 50k and TCGA GDC), including pathogenic variants and variants of uncertain significance, present different coverages and VAFs if reads were aligned to T2T-CHM13. Considering our observations, we encourage the adoption of T2T-CHM13 reference for human genomic data alignment, especially for the evaluation of clinically relevant variants. Limitations of the study This study assessed the impact of T2T-CHM13 assembly on individual features relevant to variant calling and precision medicine, both in the setting of targeted and whole genome sequencing. Although this work provides evidence for the amelioration in the detection of clinically relevant variants using T2T-CHM13 over hg38, we cannot provide a direct quantification of its practical utility for clinical diagnostics: for assessing the potential benefits of T2T-CHM13 for variant of uncertain significance re-classification or diagnosis of previously undiagnosed patients extensive ad hoc studies and validations are needed. Additionally, the differences in VAF (2.3%) and coverage (0.3%) in SNV calling for samples aligned to T2T-CHM13 compared to hg38—albeit widespread along the genome—are small, and it remains unclear whether the extent of such limited increments will have a meaningful impact on clinical interpretation. We speculated that new genome resources as T2T-CHM13 have the potential to allow the discovery of pathogenic variants underlying unsolved rare genetic diseases,31 which should, however, be tested through the analysis of tailored cohorts of patients with rare disorders. The lower median coverage of the analyzed WGS samples (30×) hindered the identification of peak shifts in VAF distributions, thereby impeding a careful assessment of RMB in the genome-wide setting. Additionally, we expect that RMB analysis of long-read data from Oxford Nanopore Technologies and PacBio sequencing might yield different trends from those observed in short-read data.32 Further limitations include the need to rely on liftOver33 conversion for comparing the same sets of genomic regions among assemblies, which does not fully account for intra-reference diversity and prevented the exploration of novel sequences added into the T2T-CHM13v2.0 assembly. Finally, potential improvements in the limit of detection resulting from using the T2T-CHM13 reference instead of hg38 should be evaluated across a larger set of samples and with orthogonal validations. In particular, the set of analyzed samples for SNV calling only incudes individuals with European descent, preventing from the assessment of the impact of reference genome change for rare variants present in the African population. Resource availability Lead contact Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Francesca Demichelis (f.demichelis@unitn.it). Materials availability This study did not generate new unique reagents. Data and code availability •Fastq, CRAM, and VCF files for cell lines and 1kGP samples were downloaded from publicly available repositories (see STAR Methods section). BAM files of 3 Mb panel sequencing (prostate cancer) and WES (breast cancer) samples are available upon request. Annotated pairwise alignment tables for ClinVar, CpG islands, and TFBS are provided on a Zenodo repository (https://doi.org/10.5281/zenodo.17953686). OpenCRAVAT annotation for ClinVar can be found among OpenCravat Modules under the name “Clinvar T2T-hg38 comparator annotator.”•The code and material for reproducing the figures is available on GitHub (https://github.com/demichelislab/ref_gen_comp). A set of scripts used for sequence analysis of UCSC tracks and homopolymers is available on GitHub (https://github.com/demichelislab/annotating_change_ref_genomes).•Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. •Fastq, CRAM, and VCF files for cell lines and 1kGP samples were downloaded from publicly available repositories (see STAR Methods section). BAM files of 3 Mb panel sequencing (prostate cancer) and WES (breast cancer) samples are available upon request. Annotated pairwise alignment tables for ClinVar, CpG islands, and TFBS are provided on a Zenodo repository (https://doi.org/10.5281/zenodo.17953686). OpenCRAVAT annotation for ClinVar can be found among OpenCravat Modules under the name “Clinvar T2T-hg38 comparator annotator.” Fastq, CRAM, and VCF files for cell lines and 1kGP samples were downloaded from publicly available repositories (see STAR Methods section). BAM files of 3 Mb panel sequencing (prostate cancer) and WES (breast cancer) samples are available upon request. Annotated pairwise alignment tables for ClinVar, CpG islands, and TFBS are provided on a Zenodo repository (https://doi.org/10.5281/zenodo.17953686). OpenCRAVAT annotation for ClinVar can be found among OpenCravat Modules under the name “Clinvar T2T-hg38 comparator annotator.” •The code and material for reproducing the figures is available on GitHub (https://github.com/demichelislab/ref_gen_comp). A set of scripts used for sequence analysis of UCSC tracks and homopolymers is available on GitHub (https://github.com/demichelislab/annotating_change_ref_genomes). The code and material for reproducing the figures is available on GitHub (https://github.com/demichelislab/ref_gen_comp). A set of scripts used for sequence analysis of UCSC tracks and homopolymers is available on GitHub (https://github.com/demichelislab/annotating_change_ref_genomes). •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Acknowledgments We thank the CIBIO Department Next-Generation Sequencing (NGS) Facility. We also thank all authors and consortia who contributed to the public datasets we queried, including the Cancer Cell Line Encyclopedia (CCLE), the T2T consortium, and the 1000 Genome Project. Funding support was provided by the Accelerator Award 2018, funded by 10.13039/501100000289Cancer Research UK (A26822) and 10.13039/100020581Fondazione AIRC per la ricerca sul cancro ETS (22792), 10.13039/501100009629Fondazione Cassa Di Risparmio Di Trento E Rovereto, and the Department of CIBIO – 10.13039/501100004004University of Trento. Y.C. is supported by the PCF Challenge Award 2024 (24CHAL12) to F.D.. F.O. was supported by an AIRC fellowship for Italy. Part of Figure 2A (https://BioRender.com/one3j7d) and Figure 3A (https://BioRender.com/ui2x0p1) were created with BioRender. Author contributions I.C., F.O., Y.C., and F.D. conceptualized and designed the study; Y.C. and F.D. supervised the study’s overall conduct; I.C. performed computational data analysis and statistical analyses; O.Q. prepared sequencing libraries; M.P. contributed materials for data analysis; F.D. was responsible for the funding acquisition for this study; I.C., F.O., Y.C., and F.D. wrote the original draft of the manuscript; I.C., F.O., O.Q., M.P., Y.C., and F.D. reviewed and edited the manuscript. Declaration of interests The authors declare no competing interests. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERDepositeddataAnnotated pairwise alignment tables for ClinVar, CpG islands and TFBSThis workhttps://doi.org/10.5281/zenodo.17953686T2T-CHM13 annotationsNurk et al.1https://github.com/marbl/CHM131kGP short-read WGS 30×Byrska-Bishop et al.14https://ftp.sra.ebi.ac.uk/vol1/fastq/CCLE short-read WGSGhandi et al.15https://www.ncbi.nlm.nih.gov/bioproject/PRJNA5233801kGP long-read ONTSchloissing et al.17https://ftp.1000genomes.ebi.ac.uk/vol1/ftp/data_collections/1KG_ONT_VIENNA/VCF files 1kGP short-read WGS 30× hg38Byrska-Bishop et al.14https://ftp.1000genomes.ebi.ac.uk/vol1/ftp/data_collections/1000G_2504_high_coverage/VCF files 1kGP short-read WGS 30× T2T-CHM13Aganezov et al.2https://s3-us-west-2.amazonaws.com/human-pangenomics/index.html?prefix=T2T/CHM13/assemblies/variants/1000_Genomes_Project/chm13v2.0/all_samples_3202/SNV calls for MSK-IMPACT 50k and TCGA GDC (2025) cohortscBioportal34https://www.cbioportal.orgencRegTfbsClustered UCSC trackENCODE Consortiumhttps://genome.ucsc.edu/cgi-bin/hgTrackUi?db=hg38&g=encRegTfbsClusteredCpG islands UCSC trackUCSC browser33https://genome.ucsc.edu/cgi-bin/hgTrackUi?g=cpgIslandExtClinVar (version 20220313) VCF (T2T-CHM13)Landrum et al.22https://s3-us-west-2.amazonaws.com/human-pangenomics/T2T/CHM13/assemblies/annotation/liftover/chm13v2.0_ClinVar20220313.vcf.gzClinVar (version 20220313) VCF (hg19 and hg38)Landrum et al.22https://ftp.ncbi.nlm.nih.gov/pub/clinvar/vcf_GRCh37/archive_2.0/2022/https://ftp.ncbi.nlm.nih.gov/pub/clinvar/vcf_GRCh38/archive_2.0/2022/dbSNP 155 VCF (T2T-CHM13)Sherry et al.35https://s3-us-west-2.amazonaws.com/human-pangenomics/T2T/CHM13/assemblies/annotation/liftover/chm13v2.0_dbSNPv155.vcf.gzdbSNP 155 VCF (hg19 and hg38)Sherry et al.35https://hgdownload.soe.ucsc.edu/gbdb/hg19/snp/https://hgdownload.soe.ucsc.edu/gbdb/hg38/snp/COSMIC Actionability signatureSondka et al.23https://cancer.sanger.ac.uk/actionabilitySoftwareandalgorithmsSamtoolsDanecek et al.27https://www.htslib.org/PicardBroad Institutehttps://broadinstitute.github.io/picard/craminoDe Costet et al.36https://github.com/wdecoster/craminoLiftOver command line toolUCSC browser33https://genome.sph.umich.edu/wiki/LiftOvertwoBitToFaUCSC browser33https://github.com/ucscGenomeBrowser/kentGATKVan der Auwera et al.37https://gatk.broadinstitute.org/bamUtilJun et al.38https://genome.sph.umich.edu/wiki/BamUtilBWALi et al.39https://github.com/lh3/bwatrimmomaticBolger et al.40https://github.com/usadellab/TrimmomaticPaCBAMValentini et al.41https://bitbucket.org/CibioBCG/pacbam/src/master/BEDtoolsQuinlan and Hall42https://bedtools.readthedocs.io/en/stable/Nextflow methylseqEwels et al.43https://nf-co.re/methylseq/2.6.0/seq-alignTurner44https://github.com/noporpoise/seq-alignStrelka2Saunders et al.26https://github.com/illumina/strelkaBCFtoolsDanecek et al.27https://samtools.github.io/bcftools/bcftools.htmlBlastnNational Center for Biotechnology Information (NCBI)https://www.ncbi.nlm.nih.gov/books/NBK279690/RsamtoolsMorgan et al.45http://bioconductor.org/packages/Rsamtools/.GvizIvanek46http://bioconductor.org/packages/Gviz/.BSgenomePagès47http://bioconductor.org/packages/BSgenome/.comparatorMarchant48https://github.com/ngmarchant/comparatorpwalignPagès49http://bioconductor.org/packages/pwalign/ggpubrThe Comprehensive R Archive Networkhttps://cran.r-project.org/web/packages/ggpubr/index.htmlgUtilsWala and Imielinski50https://github.com/mskilab-org/gUtilsBioStringsPagès et al.51https://bioconductor.org/packages/release/bioc/html/Biostrings.htmlmemesNystrom et al.52https://www.bioconductor.org/packages/release/bioc/html/memes.htmlseqvisrVragh53https://github.com/vragh/seqvisrMsCoreUtilsRainer et al.54https://www.bioconductor.org/packages/release/bioc/html/MsCoreUtils.htmlCode and material for reproducing the figuresThis workhttps://github.com/demichelislab/ref_gen_compAnalysis scriptsThis workhttps://github.com/demichelislab/annotating_change_ref_genomes REAGENT or RESOURCESOURCEIDENTIFIERDepositeddataAnnotated pairwise alignment tables for ClinVar, CpG islands and TFBSThis workhttps://doi.org/10.5281/zenodo.17953686T2T-CHM13 annotationsNurk et al.1https://github.com/marbl/CHM131kGP short-read WGS 30×Byrska-Bishop et al.14https://ftp.sra.ebi.ac.uk/vol1/fastq/CCLE short-read WGSGhandi et al.15https://www.ncbi.nlm.nih.gov/bioproject/PRJNA5233801kGP long-read ONTSchloissing et al.17https://ftp.1000genomes.ebi.ac.uk/vol1/ftp/data_collections/1KG_ONT_VIENNA/VCF files 1kGP short-read WGS 30× hg38Byrska-Bishop et al.14https://ftp.1000genomes.ebi.ac.uk/vol1/ftp/data_collections/1000G_2504_high_coverage/VCF files 1kGP short-read WGS 30× T2T-CHM13Aganezov et al.2https://s3-us-west-2.amazonaws.com/human-pangenomics/index.html?prefix=T2T/CHM13/assemblies/variants/1000_Genomes_Project/chm13v2.0/all_samples_3202/SNV calls for MSK-IMPACT 50k and TCGA GDC (2025) cohortscBioportal34https://www.cbioportal.orgencRegTfbsClustered UCSC trackENCODE Consortiumhttps://genome.ucsc.edu/cgi-bin/hgTrackUi?db=hg38&g=encRegTfbsClusteredCpG islands UCSC trackUCSC browser33https://genome.ucsc.edu/cgi-bin/hgTrackUi?g=cpgIslandExtClinVar (version 20220313) VCF (T2T-CHM13)Landrum et al.22https://s3-us-west-2.amazonaws.com/human-pangenomics/T2T/CHM13/assemblies/annotation/liftover/chm13v2.0_ClinVar20220313.vcf.gzClinVar (version 20220313) VCF (hg19 and hg38)Landrum et al.22https://ftp.ncbi.nlm.nih.gov/pub/clinvar/vcf_GRCh37/archive_2.0/2022/https://ftp.ncbi.nlm.nih.gov/pub/clinvar/vcf_GRCh38/archive_2.0/2022/dbSNP 155 VCF (T2T-CHM13)Sherry et al.35https://s3-us-west-2.amazonaws.com/human-pangenomics/T2T/CHM13/assemblies/annotation/liftover/chm13v2.0_dbSNPv155.vcf.gzdbSNP 155 VCF (hg19 and hg38)Sherry et al.35https://hgdownload.soe.ucsc.edu/gbdb/hg19/snp/https://hgdownload.soe.ucsc.edu/gbdb/hg38/snp/COSMIC Actionability signatureSondka et al.23https://cancer.sanger.ac.uk/actionabilitySoftwareandalgorithmsSamtoolsDanecek et al.27https://www.htslib.org/PicardBroad Institutehttps://broadinstitute.github.io/picard/craminoDe Costet et al.36https://github.com/wdecoster/craminoLiftOver command line toolUCSC browser33https://genome.sph.umich.edu/wiki/LiftOvertwoBitToFaUCSC browser33https://github.com/ucscGenomeBrowser/kentGATKVan der Auwera et al.37https://gatk.broadinstitute.org/bamUtilJun et al.38https://genome.sph.umich.edu/wiki/BamUtilBWALi et al.39https://github.com/lh3/bwatrimmomaticBolger et al.40https://github.com/usadellab/TrimmomaticPaCBAMValentini et al.41https://bitbucket.org/CibioBCG/pacbam/src/master/BEDtoolsQuinlan and Hall42https://bedtools.readthedocs.io/en/stable/Nextflow methylseqEwels et al.43https://nf-co.re/methylseq/2.6.0/seq-alignTurner44https://github.com/noporpoise/seq-alignStrelka2Saunders et al.26https://github.com/illumina/strelkaBCFtoolsDanecek et al.27https://samtools.github.io/bcftools/bcftools.htmlBlastnNational Center for Biotechnology Information (NCBI)https://www.ncbi.nlm.nih.gov/books/NBK279690/RsamtoolsMorgan et al.45http://bioconductor.org/packages/Rsamtools/.GvizIvanek46http://bioconductor.org/packages/Gviz/.BSgenomePagès47http://bioconductor.org/packages/BSgenome/.comparatorMarchant48https://github.com/ngmarchant/comparatorpwalignPagès49http://bioconductor.org/packages/pwalign/ggpubrThe Comprehensive R Archive Networkhttps://cran.r-project.org/web/packages/ggpubr/index.htmlgUtilsWala and Imielinski50https://github.com/mskilab-org/gUtilsBioStringsPagès et al.51https://bioconductor.org/packages/release/bioc/html/Biostrings.htmlmemesNystrom et al.52https://www.bioconductor.org/packages/release/bioc/html/memes.htmlseqvisrVragh53https://github.com/vragh/seqvisrMsCoreUtilsRainer et al.54https://www.bioconductor.org/packages/release/bioc/html/MsCoreUtils.htmlCode and material for reproducing the figuresThis workhttps://github.com/demichelislab/ref_gen_compAnalysis scriptsThis workhttps://github.com/demichelislab/annotating_change_ref_genomes Experimental model and study participant details All analyzed samples were retrieved from publicly available datasets listed in the Key Resource Table, with the exception of prostate cancer patients from the PRIME trial. The PRIME Observational Trial (ClinicalTrials.gov ID: NCT06981377) received ethics approval by Santa Chiara Hospital in Trento on 31/01/2019 (Rep. Int. # 2562) and subsequently by the Ethics Committees of all participating centers. Sequencing materials were collected from male patients with prostate adenocarcinoma with a median age of 67 (minimum and maximum age of 41 and 86, respectively) who provided informed consent in multiple trial centers located in Italy respecting regulatory standards. Evaluating the impact of sex and ethnicity was not feasible given the demographic constraints of the study population. Method details Sample collection and sequencing Breast cancer WES (96-500×) Plasma samples were obtained from a cohort of breast cancer patients enrolled on a protocol approved by the Ethics Committee of Santa Chiara Hospital in Trento (Rep.Int.12315 of July 24, 2017) with written informed consent. Eligibility criteria included breast cancer diagnosis and recommendations for neoadjuvant therapies. Blood collection and subsequent cfDNA and gDNA extraction were performed as described in a previous publication.16 For each patient, 20–50 ng of cfDNA and 100 ng of matched germline DNA (gDNA) sonicated to 180-220bp (Covaris M220) were utilized for DNA whole-exome sequencing library preparation and hybridization capture (KAPA HyperPrep Kit, SeqCap EZ HyperCap v. 2.3 Protocol; Roche). Sequencing was performed at the University of Trento by the Next Generation Sequencing Facility on Illumina HiSeq2500 platform with a paired-end protocol (100bp) with a mean coverage of 537× (282× - 731×) for cfDNA and of 96× for gDNA (89× - 109×). Prostate cancer 3 Mb panel (200-1000×) Prostate cancer plasma and buffy coat samples were obtained from patients with prostate adenocarcinoma receiving androgen deprivation therapy from PRIME Observational Trial who provided informed consent. The PRIME Observational Trial (ClinicalTrials.gov ID: NCT06981377) received ethics approval by Santa Chiara Hospital in Trento on 31/01/2019 (Rep. Int. # 2562) and subsequently by the Ethics Committees of all participating centers. Blood collection, processing, and cfDNA and gDNA extractions were performed as described in a previous publication.12 Libraries for targeted sequencing were prepared starting from 10 to 25 ng of cfDNA and 100 ng of sheared gDNA with KAPA HyperPrep Kit (Roche) following the SeqCap EZ HyperCap v2.3 protocol (for hybridization with PCF_SELECT panel v. 2.0) or KAPA HyperCap v3.0 protocol (for hybridization with PCF_SELECT v. 3.0) with a few modifications as per PCF_SELECT panel design.12 Sample sequencing was performed at the University of Trento by the Next Generation Sequencing Facility on the Illumina NovaSeq 6000 platform with 1000× and 200× intended coverage, respectively, for cfDNA and gDNA. 1kGP WGS (30×) Sample collection and sequencing protocols are described in the original publication.14 Samples data included in this study were selected randomly among all super-population groups and sex status to ensure a balanced representation of ethnicities and sex. Fastq files were downloaded from SRA EBI FTP (https://ftp.sra.ebi.ac.uk/vol1/fastq/) with accession codes provided in Table S1. CCLE WGS (30×) Sample collection and sequencing protocols are described in the original publication.15 Selected cell lines include: 22RV1 and DU145 (prostate cancer), AU565 and BT20 (breast cancer), EF021 and EF027 (ovarian cancer) and CAPAN1 and DANG (pancreatic cancer). Copy number segments were downloaded from CCLE website and coordinates with abs(Segment_Mean) < 0.5 were retained as wild-type segments. Fastq files were downloaded from SRA with accession codes provided in Table S1. 1kGP long-read WGS (30×) Sample collection and sequencing protocols are described in the original publication.17 The list of sample IDs is provided in Table S1. CRAM files were downloaded from 1KG_ONT_VIENNA repository. Sequencing data preprocessing T2T-CHM13 v2.0 reference genome and chain files were downloaded from the UCSC website. Prostate cancer 3 Mb panel (200-1000×) Adapters were removed from paired-end reads with trimmomatic40 (version 0.32). Reads were aligned with bwa mem39 (version 0.7.17-r1188) to T2T-CHM13v2.0, G1Kv37 and hg38 FASTA files. SAM files were merged with Picard (version 2.23.9) MergeSamFiles command and converted to BAM format, sorted and indexed with samtools27 (version 1.7). BAM realignment and recalibration were performed using GATK55 (version 3.8.0). Since GATK resources are not yet available for T2T-CHM13v2.0 reference genome, liftOver of hapmap_3.3.b37.vcf, 1000G_omni2.5.b37.vcf and 1000G_phase1.snps.high_confidence.b37.vcf was performed as explained in section “LiftOver of BED and VCF files”. BAM cleanup was performed with samtools calmd (version 1.7) and overlapping read pairs were clipped through bamUtil38 (version 1.0.14). SNP pileup and allelic fraction computation were run through PaCBAM41 tool (version 1.6.0), providing as input the set of all high MAF SNPs (MAF >0.2) covered by PCF_SELECT targeted panel.12 Mapping statistics were obtained with samtools (version 1.7). Breast cancer WES (96-500×), 1kGP WGS (30×) and CCLE WGS (30×) Adapters were removed from paired-end reads with trimmomatic (version 0.32). Reads were aligned with bwa mem (version 0.7.17-r1188) to T2T-CHM13v2.0 and hg38 FASTA files. SAM files were merged with Picard (version 2.23.9) MergeSamFiles command and converted to BAM format, sorted and indexed with samtools (version 1.7). BAM recalibration was performed using GATK (version 3.8.0) with previously lifted over GATK resources. SNP pileup and allelic fraction computation were run through PaCBAM tool (version 1.6.0), providing as input a set of 30 million high MAF SNPs (MAF >0.2) interspersed across chromosomes. Mapping statistics were obtained with samtools (version 1.7). 1kGP long-read WGS (30×) Since CRAM files aligned to hg38 and T2T-CHM13 were downloaded instead of fastq files, no preprocessing was applied. SNP pileup and allelic fraction computation were not performed. Mapping statistics were obtained with samtools (version 1.7) and cramino36 (version 1.1.0). Since samtools average quality estimation is not suitable for long-reads, read average quality was computed as the average from the histogram of Phred scale accuracies (corresponding to base qualities) from cramino output. LiftOver of BED and VCF files LiftOver of VCF files was achieved through Picard (version 2.23.9) LiftoverVcf function with RECOVER_SWAPPED_REF_ALT = T. LiftOver of BED files was instead performed with UCSC liftOver binary file. VCF files preprocessing and VAF computation Raw genotype calls for all 1kGP 30× samples were downloaded for hg38 and T2T-CHM13v2.0 as VCF files (description on how the files were produced is provided in the original publications2,14) from 1kGP FTP website and T2T consortium repository. Variant allele frequency (VAF) was computed from genotype calls allelic depth (AD) field as following: alt_count/(alt_count + ref_count), where ref_count and alt_count correspond to AD first and second field, respectively. VCF files were filtered for retaining only entries with at least 10 reads of coverage support and heterozygous VAF (0.2 ≤ VAF ≤0.8). Only SNPs included in the set of 30 million high MAF SNPs interspersed across chromosomes (used in section “Sequencing data processing” for SNP pileup) were kept for RMB analyses. Annotation for segmental duplications for T2T-CHM13 and non-syntenic regions between hg38 and T2T-CHM13 were downloaded from chm13 GitHub repository (https://github.com/marbl/CHM13). Theoretical VAF distribution Theoretical VAF distributions were obtained by simulating VAF values at multiple coverage bins ([5, 20], [20,30], [30,50], [50,100], [100,200]). Random sampling of allele A and allele B from a set of balanced alleles was performed for each coverage bin, maintaining a consistent per-bin SNP numerosity with respect to the hg38 patients’ WES data. Reference mapping bias analysis VAF curves were produced by selecting all heterozygous SNPs (0.2 ≤ VAF ≤0.8) with at least 20 supporting reads. For CCLE WGS pileups, only SNPs overlapping wild-type segments were considered. Distribution peaks were established by selecting the closest peak to 0.5 computed through localMaxima function from MsCoreUtils54 package. Allele swapping classification was attributed when the reference allele of hg38 assembly was equal to the alternative allele in T2T-CHM13 assembly and viceversa. SNPs with heterozygous VAF in T2T-CHM13 or hg38 correspond to calls where only one of the assemblies passes both the coverage and heterozygosity threshold. Delta VAF was computed as |T2TVAF-hg38VAF| and SNP classification was performed with respect to 0.005 threshold (delta VAF <0.005 and delta VAF ≥0.005). VAF difference among T2T-CHM13 and hg38 was tested with a one-tailed paired t test with parameter alternative = “greater” comparing T2T-CHM13 VAF distribution with hg38 VAF distribution of SNPs with delta VAF >0.005, referred throughout the text as “VAF incoherent”. Read mapping analysis Coordinates for a set of 3085 SNPs with concordant classification across at least 3 WES samples were converted to BED format (both for hg38 and T2T-CHM13 reference genome) and used to subset each WES whole blood sample’s BAM file through BEDtools42 bamtobed and intersect functions (version v2.25.0). Output QNAMEs were then given as input through READ_LIST_FILE field to Picard (version 2.23.9) FilterSamReads function to extract data related to read mapping for reads of interest from original BAM files aligned to hg38 and T2T-CHM13. SNP support was assessed based on bedtools intersect file, which matches each SNP rsID with a set of QNAMEs. Only reads mapping to a unique SNP were filtered to compare mapping features. Read classification was performed according to chromosome mapping (either multiple mapping, different chromosome, same chromosome, or uniquely mapped to either hg38 or T2T-CHM13). Multiple mapping reads were excluded from the analysis. SNP concordant support status was determined when QNAMEs assigned to the SNP were equal in hg38 and T2T-CHM13, hence SNPs with discordant support have a differential support of at least one read. SNP proximity sequence analysis The analysis of each SNP’s proximity was performed by converting its coordinates (chr and position) to a bed format specified as chromosome, position – 150 and position +150 for hg19, hg38 and T2T-CHM13 reference genomes. The FASTA sequence of each interval was obtained from Biostrings getSeq function51 giving as input the BS reference genome file (BSgenome.Hsapiens.NCBI.T2T.CHM13v2.0, BSgenome.Hsapiens.UCSC.hg19 and BSgenome.Hsapiens.UCSC.hg38) and assembly-specific genomic coordinates. The set of random SNPs was selected through createRandomRegions function of gUtils50 R package with input BSgenome.Hsapiens.NCBI.T2T.CHM13v2.0 and a GenomeRanges including all SNP coordinates, telomeres and centromeres to mask. Output ranges were lifted over to hg38 through liftOver function using hs1 (T2T-CHM13) to hg38 chain file. Similarity of DNA sequences was assessed by pairwise comparison of FASTA sequences across assemblies (T2T-CHM13 versus hg19, T2T-CHM13 versus hg38 and hg19 versus hg38) through the pwalign R package49 through the function compareSequences(). GC content was computed as the percentage of G and C bases within the analyzed sequence as:GCcontent(%)=(NC+NGNC+NG+NA+NT)∗100 Repetitiveness of DNA sequences was measured through a custom R implementation of the D2R algorithm.56 The output value represents a proxy of repetitiveness, where low values correspond to low k-mer repetition (k-mer size = 3). Since the threshold for repetitiveness detection must be defined heuristically, values were compared directly without imposing a pre-determined threshold. FASTA sequence extraction and pairwise sequence alignment FASTA sequences were obtained with twoBitToFa UCSC utility33 providing as input a specific bed file for each track and the .2bit file for the reference genome of choice. Latest version .2bit files were downloaded from UCSC golden path repositories for hg19, hg38 and hs1 reference genome (GRCh37.p13, GRCh38.p14 and T2T-CHM13v2.0 version, respectively). Pairwise sequence alignment was performed through seq-align Needleman and Wunsch algorithm implementation (https://github.com/noporpoise/seq-align) and formatted with a custom script to store the alignment result as a compact strings. Specifically, each pairwise alignment is summarized as a pipe delimited string where each event is expressed as a symbol, the event starting position and parentheses holding the number of bases undergoing the event. The symbol “?” corresponds to mismatches, “+” to insertions and “-” to deletion, assigned according to the first of the two sequences. For example, the following alignment is summarized as ?626(1)|-627(1)|+630(2).624625626627628629630631632633AAA–TACCAGAAGATA––AG 624625626627628629630631632633AAA–TACCAGAAGATA––AG In addition to the summarized string, the custom script keeps track of the alignment score produced by Needleman and Wunsch algorithm and the total number of mismatches, insertions and deletions. CpG islands annotation and methylation analysis T2T-CHM13 v2.0 UCSC CpG unmasked track was downloaded from UCSC data repository. T2T-CHM13 CpG islands bed regions were lifted over to hg19 and hg38 (see section “LiftOver of BED and VCF files”), obtaining a set of common N = 49388 CpG islands. CpG sequence analysis was performed as presented in section “FASTA sequence extraction and pairwise sequence alignment”. In addition, GC content (computed as in section “SNP proximity sequence analysis”) and the position of each C base computed for each CpG and compared among assemblies to obtain a granular classification highlighting CpGs where the number of Cs or their position varies. Reduced representation bisulphite sequencing (RRBS) fastq files from 5 CCLE cancer cell lines (PANC0504, DU4475, OAW28, MDAPCA2B and HS606T) were analyzed through Nextflow methylseq43 pipeline (version 2.6.0), both with T2T-CHM13 and hg38 as FASTA reference genomes. Bismark methylation calls for CpG context were converted to read tables through a custom script. Resulting tables were then filtered to retain only common reads and at least 4 methylation sites in hg38-aligned table. ENCODE TFBS peaks annotation and TF enrichment analysis ENCODE Transcription Factor ChIP-seq Clustered track (encRegTfbsClustered) was downloaded from UCSC data repository. encRegTfbsClusteredWithCells UCSC track for hg38 was lifted over to T2T-CHM13 and hg19 (see section “LiftOver of BED and VCF files”). TFBS peaks sequence analysis for hg19, hg38 and T2T-CHM13 was performed as presented in section “FASTA sequence extraction and pairwise sequence alignment”. The TFBS track was then filtered to retain only entries with support in at least 5 different cell lines, obtaining a set of 8 transcription factors (MYC, SP1, MAFK, CEPBP, FOXA1, FOS, JUND and CTCF). Motif enrichment analysis was performed through ame_rank() function from MEME suite R implementation52 by giving as input the whole set of sequences supporting peaks for a specific gene and the reference genome (hg19, hg38 and T2T-CHM13). The enrichment rank output was then stored for each transcription factor and compared among assemblies. ClinVar variant annotation dbSNP version 155 and ClinVar version 20220313 VCF files were downloaded from NCBI archive for GRCh37 and GRCh38 assembly. LiftOver of dbSNPv155 and ClinVar version 20220313 VCF from GRCh38 to CHM13v2.0 were downloaded from T2T-CHM13 GitHub repository. ClinVar VCF files were indexed and divided by chromosome, retaining only entries mapping to chromosomes 1:22, X and Y. Chromosomes in RefSeq accession format in dbSNP VCFs were converted through NCBI sequence report files. For comparing differences across assemblies, chromosome VCF files were merged by unique IDs, maintaining only entries present in all three VCF files. As per SNP proximal sequence analysis, each variant’s coordinates (chr and position) were converted to a bed format specified as chromosome, position – 150 and position +150 for hg19, hg38 and T2T-CHM13 reference genomes. Results are provided in Table S4. Pairwise alignment outcome was formatted and assigned to a classification based on the presence or absence of mismatches according to different thresholds.•Allele swapping: the reference base and alternative base of the variant under study are opposite among the two reference genomes;•[1,5] mismatches: among 1 and 5 mismatches;•> 5 mismatches | indels: more than 5 mismatches or presence of insertions and/or deletions (indels);Actionability classification was performed by intersecting gene names with genes included in COSMIC Actionability v1423 signature. •Allele swapping: the reference base and alternative base of the variant under study are opposite among the two reference genomes; Allele swapping: the reference base and alternative base of the variant under study are opposite among the two reference genomes; •[1,5] mismatches: among 1 and 5 mismatches; [1,5] mismatches: among 1 and 5 mismatches; •> 5 mismatches | indels: more than 5 mismatches or presence of insertions and/or deletions (indels);Actionability classification was performed by intersecting gene names with genes included in COSMIC Actionability v1423 signature. > 5 mismatches | indels: more than 5 mismatches or presence of insertions and/or deletions (indels);Actionability classification was performed by intersecting gene names with genes included in COSMIC Actionability v1423 signature. Actionability classification was performed by intersecting gene names with genes included in COSMIC Actionability v1423 signature. Actionability classification was performed by intersecting gene names with genes included in COSMIC Actionability v1423 signature. Variant calling over ClinVar sites Variant calling was performed on all tumor samples (N = 28, with N = 15 cfDNA prostate cancer samples sequenced with 3 Mb panel and N = 13 cfDNA breast cancer samples sequenced with WES) providing as input a VCF with all ClinVar sites for hg38 and T2T-CHM13 reference genome, respectively. Variant callers used for the analysis are state-of-the-art methods Strelka2 and BCFtools. Strelka2 was run with default parameters with and without Manta indel calling and –indelCandidates parameter (referred to as Strelka2.indel and Strelka2.noindel, respectively). BCFtools was run with mpileup command, specifying -d 2000 and -Q 20, and call command with -A -m parameters. Comparison of outcome calls was performed by integrating all VCF outputs based on their position and sample of origin. Variant allele frequency (VAF) was computed from genotype calls allelic depth (AD) field as following: alt_count/(alt_count + ref_count), where ref_count and alt_count correspond to AD first and second field, respectively. For the analysis of SNVs overlapping ClinVar sites in large-scale cohorts, we downloaded mutation calls from cBioPortal34 for MSK-IMPACT 50k dataset24 and TCGA GDC 2025 dataset [TCGA Research Network: https://www.cancer.gov/tcga]. We intersected mutation calls with the ClinVar sites harboring sequence change between hg38 and T2T-CHM13 identified in this study produced in this study (based on genomic position and reference/alt alleles), and computed the frequency for each variant across the cohort as the number of patients in which the variant was detected over the total number of patients of the cohorts. BLAST analysis of gene nucleotide sequences FASTA sequences of genes of interest were obtained through twoBitToFa UCSC utility33 providing as input a specific bed file for each track and the .2bit file for the reference genome of choice. The nucleotide sequence of each gene was compared with blastn57 command line tool with T2T-CHM13 as query and hg38 as subject. The output was then parsed and visualized through seqvisr53 R package. Exon coordinates for hg38 were downloaded from UCSC Table Browser33 GENCODE v47 knownGenes track. IGV-style visualization of sequencing reads BAM files were extracted with Rsamtools45 function and filtered to divide reads supporting the alternative allele from all other reads spanning the region of interest (±150 bp from the variant). Coordinates for the chromosome bands were downloaded from goldenPath website for each reference genome. Figures were built using Gviz46 R package and BSgenome47 libraries, with custom modification of Gviz functions. Comparison of read consensus with reference sequence in long-read 1kGP samples SNVs with differences in the proximal sequence among hg38 and T2T-CHM13 from Table S6 were filtered for pathogenic and likely pathogenic variants and support of VAF or coverage change in at least one sample, resulting in a total of N = 76 SNVs. Read consensus was obtained with samtools consensus (version 1.22) command by giving as input a bed with the chromosome, SNV coordinate – 150 and SNV coordinate +150 separately for hg38 and T2T-CHM13 aligned samples and with -q parameter equal to 20. The consensus sequence for each SNV in each sample was then directly compared to the reference genome of choice with a custom R script. Specifically, Levenshtein similarity was computed through comparator48 R package among the FASTA sequence consensus and FASTA sequence extracted from BSgenome47 libraries with the same bed coordinates. Quantification and statistical analysis All statistical analyses were performed in R (version 4.4.1). Paired Wilcoxon tests reported in boxplots (both one-tailed and two-tailed) were performed through the ggpubr R package using the function stat_compare_means(). Paired one-tailed Wilcoxon tests for VAF T2T-CHM13 > VAF hg38 in the RMB analysis barplots were performed in R through the wilcox.test() function. Enrichment for variation in homopolymeric regions was evaluated in R with the chisq.test() function. For each statical test we have considered changes significant if p values were below 0.05. Details for each statistical test are present either in the text or in the figure legends. Additional information on how the p values were included in the figures and how figures were generated can be retrieved from the GitHub repository https://github.com/demichelislab/ref_gen_comp.