Beyond SNVs: What AACR 2026 Revealed About the Future of Genomic Variant Detection

Written By Ana Diaz

For years, “genomic profiling” meant single-nucleotide variants, a list of hotspot mutations, a panel, and a VAF. At AACR 2026 in San Diego this April, that was no longer the norm, and structural variants, extrachromosomal DNA, and the 3D genome began to take center stage in many conversations.   

Structural variants have gone mainstream

Structural variants have always been the harder sell; they don't fit neatly into a hotspot panel. Not this year. Across posters, oral sessions, and the exhibit floor, SVs were treated as central to understanding a tumor genome, not a specialty add-on.  

Dovetail brought three posters of its own to San Diego, plus three more from academic collaborators building on Dovetail's Linked-Read chemistry: 

Together, they made the case from three angles. Against an independent benchmark, a new SV caller called Dovetail Precise recalled 90% of the Genome in a Bottle Consortium's draft somatic SV truthset (HG008-T, a pancreatic cancer tumor-normal pair), versus 18% for a standard contact-map-based caller, with near base-pair breakpoint precision. In a genuinely complex cancer genome, a Sylvester Myeloma Institute team profiled four relapsed myeloma patient-derived xenografts: even at 30x tumor-only coverage, LinkPrep's SV calls carried 10–100x more read support than 80x/30x WGS, 15x PacBio, or 400x optical genome mapping, recalled 94.2% of the shotgun SNV/indel truth set, and traced neo-chromosome structures back to IGH and IRF4 enhancer-hijacking events. A companion poster went further, generating the first chromosome-scale haplotype-resolved myeloma genomes (94.3% phasing per autosome, haplotypes up to 247.8 Mb) to map how rearrangements like t(11;14) reorganize regulatory activity around genes like CCND1.

On the samples labs actually have, a prostate cancer poster used linked-read contact patterns to distinguish ecDNA from HSRs, then the University of Michigan's Center for Translational Pathology (led by Arul M. Chinnaiyan) applied the same method to 20 prostate FFPE biopsies and called AR-associated ecDNA correctly in 3 of 3 validated cases. That FFPE result got the most attention at the booth all week. 

The fourth angle came from Standford’s Majeti lab, looking at AML cases with no canonical driver at all. In two karyotypically normal patients, the same linked-read approach found cryptic rearrangements standard panels missed (a t(2;13) translocation and 10 Mb chr2 inversion in one, a 75 Mb chr8 inversion in the other) each confirmed to base pair breakpoints by PCR and Sanger sequencing. Overlaying 3D contact maps on matched RNA-seq showed these breakpoints creating neoloops that repositioned enhancers near SOX17, POU4F1, and TGS1 suggesting enhancer hijacking as a hidden driver in “driverless” AML.  

Taken together, the four data sets argue that current AML and cancer genomics may be systemically undercounting structural variants, and that a meaningful fraction of “driverless” cases may in fact be driven by rearrangements and 3D genome disruptions that standard panels and karyotyping simply can’t see.  

What is read support and why does it matter?

Read support is the number of independent reads that confirm a breakpoint; the more reads backing a call, the more confident we can be that it's real rather than an artifact. Standard short-read WGS can only draw on reads within about 2 kb of a breakpoint; Dovetail's linked reads use breakpoint-spanning and breakpoint-flanking reads to extend that window beyond 100 Mb, backing each rearrangement with a whole cloud of evidence instead of a handful of fragments. That's why researchers who've relied on WGS for years keep asking the same question: what would higher read support add to a call I already have?  

Dovetail's linked-reads generate an order of magnitude more physical evidence per structural variant than PacBio HiFi or standard Illumina shotgun sequencing. Averaged across large SVs (cis rearrangements over 1Mb and trans rearrangements, n=22): 

That's roughly 100x higher average read support, and linked-reads remained sensitive at coverage and purity levels where HiFi and shotgun sequencing fell off a cliff. The myeloma cohort independently confirmed this, with 30x tumor-only LinkPrep outperforming every orthogonal technology by 10-100x per call. That level of read support means more confidence in every call - both the ones already found and the ones still waiting to be discovered. 

ecDNA: the amplification mechanism hiding in plain sight

Extrachromosomal DNA (ecDNA) was a recurring topic. ecDNA are circular DNA fragments that carry oncogenes outside the normal chromosome structure, driving the extreme, focal copy number amplifications common in aggressive cancers. 

Dovetail's prostate cancer poster identifies ecDNA by contact-frequency decay: ecDNA amplicons show a spike in long-range interactions without the distance-decay pattern typical of structural variants, whereas HSRs show elevated copy number without that long-range signal. After validating the approach in COLO320 cell lines, the team applied it to mCRPC samples. One, MI-2, showed a strong ecDNA signal (max Z-score 45.3, detectable even at 5x coverage), later confirmed by AmpliconArchitect reconstruction of a ~3.6 Mb circular element carrying roughly 33 copies of AR. That capability now lives within Variant Analysis on the Dovetail Analysis Portal (DAP).  

The distinction matters clinically: AR amplification is a well-documented driver of resistance to androgen-deprivation therapy, and ecDNA is increasingly understood to be its vehicle. Whether an amplification resides on ecDNA or is locked into a chromosome as an HSR shapes how it's likely to behave under treatment. Standard copy number data can't tell you this, but this workflow can. With the Michigan team validating the same approach on FFPE tissue at 3-for-3 accuracy, that answer is now available from archival samples, not just fresh-frozen research material.  

FFPE, finally solved

FFPE tissue is how most of the world's tumor tissue is stored, and it's also one of the most challenging sample types in genomics. Formalin fixation crosslinks DNA, fragments it into short pieces, and introduces deamination artifacts that mimic real mutations - this is close to a worst-case scenario for SV detection, which depends on intact, long-range genomic information. Long-read platforms typically require high-molecular-weight DNA that FFPE can't reliably provide; standard short-read WGS is left to catch large rearrangements with reads that barely span 2 kb around a breakpoint.  

Dovetail's FFPE assay closes that gap, and its official commercial launch at this year's meeting, marking the transition from early access to full availability, was one of the show's biggest product moments. The workflow is built around degraded FFPE input from the start: samples are de-paraffinized, chromatin is conditioned in situ, fragmented with MNase, and religated via proximity ligation into standard linked-read libraries. Because that ligation captures which fragments started out near each other before formalin ever touched them, it reconstructs long-range linkage that a same-length shotgun library simply can't recover. One FFPE library delivers structural variants, SNVs, indels, and CNVs together, validated across tumor types at 10x to 30x coverage, with high concordance to matched fresh-frozen tissue.  

The TAT for data analysis is what?!

For many researchers, data analysis has long been a major barrier to adopting Hi-C technologies. Without a bioinformatician familiar with Hi-C analysis, learning specialized command-line tools and navigating complex pipelines can be overwhelming. Even for experienced computational users, processing datasets and generating interpretable results often take days or even weeks, slowing the pace of discovery. With this challenge in mind, the team at Dovetail developed the Dovetail Analysis Portal to simplify structural variant analysis from start to finish. From raw FASTQ files to a comprehensive structural variant report, the portal's automated pipeline delivers results in as little as one day. It’s as simple as uploading FASTQ files and clicking "Analyze.” No command-line interfaces, complex pipelines, or bioinformatics expertise required.  

Where this leaves the field

Hotspot SNV detection remains the backbone of clinical genomics, and nothing at AACR 2026 challenged that. But between the formal program and a week of booth conversations, the direction was clear: structural variants, ecDNA, and 3D genome context are no longer specialty add-ons. They're becoming table stakes for any serious genomics platform, exactly the ground Dovetail's linked-read technology was built to own. 

Sources: AACR Annual Meeting 2026 program and abstracts (aacr.org, aacrjournals.org/cancerres); AACR daily blog recap, April 22, 2026; Dovetail Genomics AACR 2026 poster program and FFPE/Dovetail Analysis Portal product pages (cantatabio.com); Dovetail Genomics FFPE assay press materials (BioSpace, PR Newswire); Inocras and Guardant Health AACR 2026 presentation summaries (investor relations pages, BioSpace); GeneCentric AACR 2026 press coverage (BioSpace); OncLive AACR 2026 preview coverage; Frontiers in Oncology (2026), “Long-read sequencing for cancer liquid biopsy: advancing precision oncology.”

Ana Diaz