Journal Club: cfDNA Detection, SV Calling, and Driver Gene Prediction in Cancer Genomics

· 9 min read · Paper Feed

Journal Club: cfDNA Detection, SV Calling, and Driver Gene Prediction in Cancer Genomics

Conceptual illustration: cfDNA Detection, SV Calling, and Driver Gene Prediction in Cancer Genomics

Introduction

Cancer is a genetic disease, but reading a tumor genome accurately is still hard. Circulating cell-free DNA (ccfDNA) often sits at vanishingly low fractions, tumor specimens come admixed with normal stroma, and structural variants (SVs) routinely span breakpoints that short reads cannot resolve. On top of that, downstream interpretation only matters if detected variants can be tied back to the canonical driver genes and pathways that actually drive proliferation, immune evasion, and therapy resistance.

This journal club covers three 2025 papers that together span modern cancer genomics from bench to bedside. Cheng et al. (Nature Methods) push the wet-lab side by combining error-corrected duplex sequencing with cost-effective flow-based WGS to profile ccfDNA at part-per-million sensitivity. Kerem Aydin et al. (Scientific Reports) take the computational side, benchmarking eight long-read SV callers and tool combinations on paired tumor–normal cancer cell lines. Ubaid et al. (Cancer Genetics) zoom out with a multi-omics review of oncogenic determinants, giving a biological frame for interpreting any newly called variant.

What ties them together is a shared observation: reliable detection of low-prevalence, biologically meaningful alterations depends on chemistry, software, and curated knowledge bases all at once. A weak layer anywhere caps the whole pipeline.

Paper 1: Duplex-Corrected, Low-Cost WGS for Liquid Biopsy

Key Findings

  • An error rate of 7.7 × 10⁻⁷ was reached using duplex error correction on Ultima Genomics flow-based WGS, pulling the false-positive ceiling down for low-frequency variant calling.
  • Deep ~120× coverage on this lower-cost platform allowed detection of circulating tumor DNA in the parts-per-million range, comparable in sensitivity to far more expensive short-read platforms.
  • The authors demonstrated tumor-informed ctDNA tracking and a tumor-naïve application, profiling disease burden in melanoma and urothelial cancer patients without requiring a matched tumor biopsy.

Methodology & Study Design

The team used Ultima Genomics' high-throughput, low-cost flow-based sequencing chemistry (which has much lower reagent costs per base than legacy Illumina platforms) and layered on duplex error correction, a strategy in which both strands of each original DNA duplex are tagged and sequenced independently, so true variants appear on both strands while PCR or sequencing errors do not. Applied to ccfDNA from cancer patients, this is the first demonstration of duplex-corrected WGS at whole-genome scale on a non-Illumina platform, with error rates low enough for ultra-sensitive variant calling. The pipeline enables direct whole-genome signal integration for ctDNA quantification, a conceptual alternative to targeted mutation panels.

Significance

Earlier duplex-corrected WGS work (e.g., the original "Duplex-Seq" approach) was too expensive to scale. By pairing duplex correction with a fundamentally cheaper sequencing chemistry, Cheng et al. make population-scale, ultra-sensitive WGS economically tractable, including ctDNA-based cancer monitoring and studies of somatic mosaicism in aging and disease.

Paper 2: Benchmarking Long-Read SV Callers on Cancer Genomes

Key Findings

  • Eight widely used SV callers were evaluated on paired tumor/normal data from the NCI-H2009 lung cancer cell line and the well-characterized COLO829 melanoma cell line.
  • No single caller dominated across all SV types; performance varied by variant class (insertions, deletions, inversions, translocations, copy-number-associated rearrangements).
  • Tool combinations consistently outperformed individual callers, identifying the highest-confidence set of somatic SVs after VCF merging and matched-normal subtraction.

Methodology & Study Design

Long-read sequencing (PacBio / Oxford Nanopore) is attractive for SV detection because reads span repetitive regions and complex breakpoints where short reads collapse. The authors ran each of eight callers, including Sniffles, cuteSV, SVIM, NanoSV, pbsv, and others, independently on tumor and matched-normal long reads, then merged calls and subtracted germline events to flag candidate somatic SVs. The COLO829 truth set (with curated somatic SVs from the SEQC2 consortium) provided a high-confidence evaluation ground, while NCI-H2009 added biological diversity.

Significance

Long-read somatic SV calling has grown fast, but rigorous benchmarking has lagged tool development. This study delivers a practical framework: use multiple callers, take intersections or weighted unions, and always subtract the matched normal. It also exposes where current long-read SV tools still fall short, particularly for insertions and certain translocations, which is useful guidance for future tool development.

Notable coincidence: COLO829 melanoma DNA appears in both Papers 1 and 2. Cheng et al. use it for benchmarking their duplex-WGS pipeline; Kerem Aydin et al. use it as the truth set for somatic SVs. The cell line is quietly turning into a community reference standard.

Paper 3: Oncogenic Determinants Across Tumor Types via Multi-Omics Integration

Key Findings

  • Provides a structured synthesis of major oncogenes (KRAS, PIK3CA, MYC, EGFR) and tumor suppressors (TP53, PTEN, CDKN2A, RB1), with mechanistic detail on each.
  • Highlights the growing importance of chromatin remodelers (ARID1A, KMT2D) as tumor suppressors, historically under-appreciated relative to canonical drivers.
  • Discusses emerging therapeutic strategies including synthetic lethality, pathway-targeted inhibitors, and biomarker-guided precision oncology, supported by resources such as the IntOGen driver-gene prioritization platform.

Methodology & Study Design

This is a narrative review synthesizing insights from large-scale cancer genome projects (TCGA, ICGC, PCAWG), whole-genome sequencing studies, and multi-omics integration across genomic, transcriptomic, proteomic, and epigenomic layers. The authors organize findings around hallmark cancer pathways: cell cycle, apoptosis, metabolic reprogramming, immune evasion, providing a conceptual scaffold for the technical detection work described in Papers 1 and 2.

Significance

Detection without interpretation is a dead end. Ubaid et al. argue that the future of cancer genomics lies not in any one data type but in their systematic integration. Variants detected by Cheng et al.'s pipeline, or the somatic SVs called by Kerem Aydin et al.'s tool combinations, ultimately need to be mapped onto curated biological frameworks to inform treatment decisions.

Synthesis & Discussion

Read together, these three papers trace a vertical slice through modern cancer genomics: wet-lab chemistry → computational detection → biological interpretation.

A coherent pipeline emerges:

  1. Detection (Paper 1) generates high-confidence somatic SNVs/indels from ccfDNA at single-molecule precision.
  2. Structural analysis (Paper 2) complements this with SVs that may span the same locus, or hit entirely different pathways (e.g., enhancer hijacking, gene fusions).
  3. Interpretation (Paper 3) maps any called variant onto curated driver-gene and pathway knowledge to predict phenotype and therapeutic vulnerability.

Convergences and tensions:

  • Both Papers 1 and 2 lean on increasingly cheap sequencing, but only Paper 1 has fully embraced the cost-saving Ultima platform for duplex correction; long-read SV calling in Paper 2 remains expensive.
  • Paper 2's finding that no single tool wins echoes a recurring lesson in bioinformatics, and reinforces the case for tool combinations that Paper 1's downstream variant calling would also benefit from.
  • Paper 3's emphasis on chromatin remodelers like ARID1A and KMT2D connects directly to Paper 2: such genes are often disrupted by SVs (large deletions, translocations), making long-read detection particularly valuable.

Open questions:

  • Can duplex-corrected WGS workflows be extended to long reads, marrying the strengths of Papers 1 and 2?
  • Is the COLO829 truth set robust enough to serve as a community-wide somatic SV benchmark, analogous to GIAB's role in germline variants?
  • How do we systematically reconcile somatic SV calls with the chromatin-remodeling and non-coding driver landscape Ubaid et al. highlight?

For Your Lab Meeting

  1. Cheng et al. rely on a single chemistry (Ultima). Would duplex error correction translate equally well to PacBio HiFi or Oxford Nanopore duplex protocols, or are platform-specific error profiles too idiosyncratic?
  2. Tool combinations won in Paper 2, but at what merging threshold? How sensitive are their conclusions to changes in VCF merging and subtraction parameters, and how can we mitigate that brittleness?
  3. Paper 3 emphasizes ARID1A and KMT2D as under-recognized drivers. Are these typically detected through SNVs only, or do SVs (Paper 2) and chromatin accessibility (ATAC-seq, Hi-C) reveal additional contexts worth incorporating?
  4. Tumor-naïve ctDNA detection (Paper 1) is striking, but how does it scale at population level, and what are the bioinformatic challenges of building a general-purpose "tumor-of-origin" classifier on top of duplex WGS?
  5. Where will the next bottleneck emerge? If chemistry (Paper 1), detection (Paper 2), and interpretation (Paper 3) all improve simultaneously, is variant validation and functional characterization the new rate-limiting step?

Key Terms

  • Duplex sequencing / duplex error correction — A protocol that tags both strands of the original double-stranded DNA molecule with unique molecular identifiers and requires a variant to be observed on both strands before being called, dramatically reducing sequencing errors.
  • Circulating tumor DNA (ctDNA) — Tumor-derived fragments of cell-free DNA released into the bloodstream; used as a minimally invasive "liquid biopsy" for cancer monitoring.
  • Somatic structural variant (SV) — A genomic rearrangement (deletion, duplication, inversion, translocation, or complex event) acquired in a tumor but not present in the patient's germline.
  • Multi-omics integration — The combined computational analysis of multiple high-throughput data layers (genomics, transcriptomics, proteomics, epigenomics, metabolomics) to derive insights not visible in any single layer.
  • Synthetic lethality — A genetic interaction in which perturbation of either of two genes individually is tolerated, but simultaneous perturbation causes cell death, exploited therapeutically (e.g., PARP inhibitors in BRCA-deficient tumors).

References

  1. A. P. Cheng, Adam J. Widman, Anushri Arora, I. Rusinek, Aaron Sossin, Srinivas Rajagopalan, et al. (2025). Error-corrected flow-based sequencing at whole genome scale and its application to circulating cell-free DNA profiling. Nature Methods. https://doi.org/10.1038/s41592-025-02648-9

  2. Safa Kerem Aydin, Kübra Yılmaz, & A. Acar (2025). Benchmarking long-read structural variant calling tools and combinations for detecting somatic variants in cancer genomes. Scientific Reports. https://doi.org/10.1038/s41598-025-92750-x

  3. Saba Ubaid, R. Kushwaha, M. Kashif, & Vivek Singh (2025). Comprehensive analysis of oncogenic determinants across tumor types via multi-omics integration.. Cancer Genetics. https://doi.org/10.1016/j.cancergen.2025.08.010

Cancer Genomics Oncology Tumor