henhaobio
Sign in

bioRxivGenomicsPreprint

Single-Cell Indel Detection Enhances Genetic Ancestry and Cellular Lineage Analysis

Ziyi Wang, Ken CheniD, Jinzhuang DouiD

PDFDOICodeData
In-depth full-text digest available · Jump to digest ↓

Abstract

Small insertions and deletions (Indels) provide critical information for cancer genomics and clonal evolution, yet their detection from single-cell sequencing (SCS) such as scRNA-seq and scATAC-seq remains challenging due to sparse coverage, alignment artifacts, and RNA editing. Here, we present Monopogen-Indel, a bioinformatics framework for accurate germline and somatic Indel detection through haplotype-aware variant calling, dynamic template matching in repetitive regions, and cell-population- based allele segregation analysis. We validated germline Indel detection in human retina snRNA-seq with matched bulk whole genome sequencing (WGS). Monopogen-Indel detected 41,000-45,000 germline Indels per sample, with >70% precision and >90% genotyping accuracy. Using 65 heart left ventricle snATAC-seq samples, indel-based global ancestry inference segregated genetic ancestry comparably to SNVs, establishing indels as an independent marker of genetic diversity in SCS. In scRNA-seq from 43,717 cells across four anatomic sites of a patient with high grade serous ovarian cancer (HGSOC), Monopogen- Indel identified ~50,000 germline Indels per sample at 86% WGS-validated precision and an average of 1,040 de novo Indels per sample. Approximately 50% of de novo Indels reflected biological sources, including variants WGS variants, RNA editing, and cell-type-specific patterns. Somatic SNVs were correctly restricted to CD45- cells, indicating high specificity in delineating somatic from germline variants. In summary, Monopogen-indel is the first framework dedicated to indel detection from SCS and expands the utility of existing single-cell data for population genetics and cancer evolution. The module is integrated into Monopogen repository https://github.com/KChen-lab/Monopogen.

In-depth summary

Claude Opus 5.5 · xhigh effort · 7 Nov 2026 · from the full text of version 1 (7 Sept 2026) · CC BY 4.0

Monopogen-Indel calls germline indels from single-cell and single-nucleus RNA data at 73.0% (retina) and ~87% (ovarian cancer) WGS-validated precision, though no single-cell somatic indel matched a WGS somatic indel.

Single-cell somatic mutation callers such as Monovar, SComatic and Monopogen have concentrated on SNVs, leaving indels, which can trace clones and disrupt genes, little studied because of uneven single-cell coverage, homopolymer errors and misalignment. This work adds indel calling to Monopogen, combining reference-panel linkage disequilibrium (LD) for germline genotyping, repeat-aware read templates and cell-population allele segregation to separate somatic from germline calls.

Key results

  1. ShownIn four human retina snRNA-seq samples, germline indel calls were moderately precise and their genotypes mostly matched WGS, but recall was very low.

    Mean 42.6k (41,016-45,734) germline indels per sample; on average 31,145 overlapped a WGS gold standard (at least 4 supporting reads), giving 73.0% precision and 3.93% recall in callable regions, vs ~20% recall for SNVs in the same sample. Genotype concordance was >90% in all four samples; most errors were homozygous/heterozygous swaps. · Validation of Germline Indel Detection Accuracy; Performance Metrics; Genotyping Accuracy Assessment

  2. ShownIn scRNA-seq from one high-grade serous ovarian cancer (HGSOC) patient across 4 anatomic sites, germline indel calls were ~87% precise but trailed Monopogen SNV calls on both precision and recall.

    Mean 56,465 (52,080-60,248) germline indels per dataset at ~87% precision vs 97% for SNVs; mean recall 11.31% vs 28.82%. Variant counts did not differ significantly across sites or cell types. The abstract instead reports ~50,000 indels per sample at 86% precision. · Validation of Germline Indel Detection Accuracy; Abstract

  3. SuggestedGermline indels called from snATAC-seq reproduced the SNP-based ancestry groupings of 65 heart donors.

    ENCODE heart left ventricle snATAC-seq, 65 donors: PCA of 1,249,093 germline indels recovered the ancestry groups (African, Central/South Asian, East Asian, European) assigned by LASER projection of Monopogen SNP genotypes onto HGDP. Agreement is shown only visually (Figure 2); no concordance metric is reported. · Global Ancestry Inference with germline Indels…; Global Ancestry Analysis; Figure legends

  4. SuggestedSomatic SNV calls were specific to tumor cells, but no single-cell somatic indel was confirmed by WGS.

    Tumor vs peripheral blood mononuclear cell (PBMC) WGS (bidirectional Mutect2) gave 2,420 somatic SNVs and 300 somatic indels. Across 43,717 cells, a mean of 1,040 indels (684-1,239) was called per site; none overlapped WGS somatic indels. All SNVs matching WGS tumor mutations came from non-CD45+ cells, not CD45+ immune cells. Precision: 7.62% for SNVs vs 2.59% for indels. · Somatic Indel Detection Performance from SCS…; Somatic Variant Calling Evaluation

  5. SuggestedAbout half of de novo indels could be traced to biological or WGS-supported sources, a smaller share than for de novo SNVs.

    Of 8,099 de novo indels and 58,096 de novo SNVs, 15.93% of somatic indels vs 32.76% of somatic SNVs recurred in more than 2 sites. WGS pileups, RNA-editing databases, RepeatMasker and cell-type enrichment (Fisher's exact test) explained ~55% of de novo indels vs ~79% of SNVs. In the examples shown, indels were tumor-restricted (absent from lymphocytes); some also appeared in fibroblasts. · Characterization of de novo Somatic Variants; De Novo Variant Characterization

Methods, data and reproducibility

Extends Monopogen to indels under 110 bp: pseudobulk candidates are genotyped and phased with 1000 Genomes Phase 3 (1KG3) LD; non-panel candidates are assigned to cells by CTPMScan (Concurrent Two-Pointer Motif Scan) template matching with repeat-extended flanks, then filtered by a support vector machine (SVM) and a cell co-segregation model. Calls are checked against matched WGS.

  • Retina snRNA-seq with matched bulk WGS (Human Cell Atlas) · human · n = 4 samples · Human Cell Atlas project f0f89c14-7460-4bab-9d42-22228a91f185
  • MSK-SPECTRUM high-grade serous ovarian cancer scRNA-seq, one patient (Donor 115), 4 anatomic sites, cells stratified by CD45 status · human · n = 43,717 cells · Synapse syn25569736; dbGaP phs002857.v3.p1
  • Matched WGS of primary tumor and PBMC from MSK-SPECTRUM Donor 115, used as germline/somatic ground truth · human · n = 1 donor · dbGaP phs002857.v3.p1
  • ENCODE heart left ventricle snATAC-seq · human · n = 65 donors
  • 1000 Genomes Phase 3 (1KG3) haplotype reference panel for germline genotyping and phasing · human (1000 Genomes Project panel)
  • Human Genome Diversity Project (HGDP) genotype panel for ancestry projection · human · n = 938 individuals, 53 populations, 632 958 SNVs
Methods & tools
Monopogen-Indel, Monopogen, Samtools, BCFtools, LASER (Trace module), BWA-MEM 0.7.18-r1243-dirty, GATK v4.6.1.0, Mutect2, RepeatMasker, majiq3, 1000 Genomes Phase 3 reference panel, HGDP reference panel
Software & environment
Versions are given only for the WGS processing: BWA-MEM 0.7.18-r1243-dirty and GATK v4.6.1.0 (GRCh38). No versions, container or requirements file are given for Monopogen-Indel itself.
Reproducibility gaps
Raw MSK-SPECTRUM sequencing files are controlled access (dbGaP phs002857.v3.p1). The SVM's features, training data and hyperparameters, and the cell-population model, are only outlined. No compute requirements are reported.
Limitations (6)
  • Short indels (under 5 bp) are prone to PCR slippage in homopolymers, sequencing error and misalignment. Sparse per-cell coverage and allelic dropout (scRNA-seq) and Tn5 bias (scATAC-seq) add noise, and residual errors likely persist in repeats despite RepeatMasker and RNA-editing filtering. · stated by the authors
  • Lower tumor purity can shift LD patterns and weaken somatic detection, probably more in scRNA-seq than in scATAC-seq. Germline variants with allele-specific expression may be miscalled as somatic, especially in low-purity or allelically imbalanced tumors. · stated by the authors
  • Telling true de novo indels from artifacts remains hard: some low-allele-frequency indels in low-complexity regions may be technical. Orthogonal validation (targeted single-cell DNA sequencing, long reads) is left for future work. · stated by the authors
  • Germline genotyping uses only the 1KG3 panel; the authors plan to add TOPMed, All of Us and UK Biobank panels to account for ethnic disparity. · stated by the authors
  • Somatic indel calling is not validated against ground truth: in the single HGSOC patient, no single-cell somatic indel matched any of the 300 WGS somatic indels and indel precision was 2.59%, so the somatic claims rest on indirect categorization. · noted in this summary
  • Numbers differ between sections (germline indels ~50,000 at 86% in the abstract vs 56,465 at ~87% in results; de novo indels explained ~50% vs ~55%; the same '27' is given both as somatic calls and as WGS-germline overlaps). No other indel caller is compared, and ancestry agreement is judged visually. · noted in this summary

Read this if you want to pull germline or somatic indels, or indel-based ancestry, out of existing scRNA-seq, snRNA-seq or snATAC-seq data and need realistic precision and recall figures against matched WGS.

Citations
0

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISLuo '26Dou '22Curnin '19Fang '16Prashant '21Gusev '21
63% semantic matchBioinformatics
Rui Luo, Ziyi Wang, Jinzhuang Dou+8 more·bioRxiv·Jun 14, 2026