ASHG 2026: Five Trends to Watch in Rare Disease Diagnostics
Sookjin Lee
Chief Business Officer (CBO) | Ph.D.
A business expert leading innovation in precision medicine and genomics for 20 years. Accelerating the commercialization of innovative technologies with academic expertise and market insight.
Summary
This year’s ASHG made one thing clear: rare disease diagnostics is shifting from “reading more” to “interpreting what we have already read, more deeply and over a longer horizon.”
The 3billion research team attended the ASHG (American Society of Human Genetics) Annual Meeting, held in Montréal, Canada, in October 2026. Below we share the trends we found most relevant to clinical practice. Rather than focusing on individual figures, we focus on what each trend changes for patient care.
The key messages for clinical practice come down to three points:
- A negative result is not the end. More and more answers are being found through reanalysis and next-step testing.
- Testing is converging. A single test is increasingly able to capture more variant types, and even methylation.
- Interpretation remains the work of experts. AI is a powerful tool, but final judgment and counseling must stay with experts.
1. Long-read sequencing: many tests in one
Long-read sequencing is moving from a research tool into clinical testing. The key point is not so much “a more accurate test” as “the consolidation of many tests.”
What was presented
- Long-read data from large populations continue to be released, and gnomAD, the most widely used reference control for variant interpretation, has also released long-read data.
- Radboudumc and Maastricht UMC+ in the Netherlands compared long-read sequencing side by side with standard-of-care testing in about 800 patients. It reproduced existing results almost entirely, added further diagnoses, and was reported to be able to replace several separate tests (published in NEJM).
- Several US laboratories validated workflows that assess SNVs, CNVs, structural variants, repeat sizes, and methylation from a single genome test. Conditions that currently require separate tests, such as Fragile X, Prader-Willi/Angelman, and FSHD, were also evaluated from the same data.
- Multiple centers repeatedly showed that long-read sequencing yields additional diagnoses in families who remained unsolved after conventional testing.
Why it matters
- For patients, it means less time spent going through tests one after another and waiting for each result.
- It reaches regions that short-read sequencing structurally misses (complex structural variants, repeats, pseudogenes), and it can determine whether two variants sit on the same allele (phasing) without parental testing.
- As large long-read population datasets (UK Biobank, All of Us, and others) accumulate and a new reference genome (Human Pangenome 1.0, expected in 2027–28) is prepared, the foundation for interpreting newly found variants is rapidly being built.
What to expect in the clinic
- For now, long-read sequencing will mainly serve as a next-step test for WES/WGS-negative patients, and as costs fall it will become a candidate first-tier test.
- An improved reference genome is expected to recover more variants even from existing short-read WGS data. Analysis of existing short-read data and reanalysis of stored samples are therefore also expected to benefit.
2. Episignatures: looking at variants and methylation together
An episignature is a blood DNA methylation pattern shared by patients with a specific genetic disorder. Episignatures are now moving toward being read directly from sequencing data, without a separate test.
What was presented
- The Hospital for Sick Children (SickKids) in Toronto applied an episignature classifier built for methylation arrays to long-read data and obtained the same calls as the array.
- Cases were reported in which long-read sequencing detected a signature in WES/WGS-negative patients, and the causal deep intronic variant was then found in the same data.
- When episignatures identified the cause in undiagnosed patients with developmental and epileptic encephalopathy (DEE), the causal variants were already present in the existing WES data.
Why it matters
- For VUS in chromatin regulator genes such as KMT2D and ARID1B, methylation patterns provide functional evidence.
- A single test can answer both “which variant is present?” and “did that variant actually change function?”
What to expect in the clinic, and limitations
- It opens a new path to reclassifying VUS that are otherwise difficult to interpret.
- For now, only conditions with a known signature can be assessed. Many genes have no signature, so a negative episignature result cannot be used as evidence for a benign classification.
- Each time new patterns are added to classifiers, reanalysis may yield additional diagnoses.
3. Reanalysis: a negative result is not the end
Once again this year, one of the most frequently repeated messages was that “the answer is often already in the data.” A negative result is “negative by the standards at the time of testing.”
What was presented
- Phenotype-tailored strategies (Care4Rare Canada): Undiagnosed families were divided into four groups by phenotype, with a different strategy for each group. For example, for patients with a clear clinical diagnosis but negative testing, mosaicism was suspected and high-depth targeted testing and non-blood samples (buccal cells, lesional tissue) were used, leading to a diagnosis in 49%.
- Revisiting with new tools (Keio University, Japan): Existing WGS data were reanalyzed with tools dedicated to mobile element insertions (MEIs), yielding new diagnoses. The findings also raised the possibility that some of these are founder variants passed down from a common East Asian ancestor.
- Broadening the transcript view (Boston Children’s Hospital): Looking only at the representative (canonical) transcript systematically misses pathogenic variants in tissue-specific transcripts.
- Newly recognized disease genes: The value of genotype-first reanalysis, revisiting existing data in light of recently discovered disease genes such as the non-coding RNA gene RNU4-2, was emphasized.
- Targeted RNA testing: Rather than as a standalone diagnostic test, RNA-seq was most effective for VUS reclassification when used to confirm splicing abnormalities after DNA testing had narrowed down the candidates.
Why it matters
- Because new disease genes, new analysis tools, and improved interpretation standards keep emerging, the same data can yield a different answer over time.
- Diagnostic opportunities can be reopened without an additional blood draw or cost burden for the patient.
What to expect in the clinic
- For negative patients, consider periodic reanalysis, and revisit the data promptly when new symptoms or family history information emerge.
- If a specific condition is strongly suspected clinically but testing is negative, follow-up strategies such as sample selection that accounts for mosaicism, or RNA analysis, can help.
4. AI in variant interpretation: calibrated tools used alongside experts
AI is becoming a standard tool in variant interpretation. This year’s focus was not “AI makes the call” but “AI supports expert judgment within validated standards.”
What was presented
- Calibrating ACMG criteria with data: An approach was presented that keeps the ACMG rules unchanged but uses machine learning (ML) to calibrate, gene by gene, the strength at which each rule is applied. A substantial share of VUS could be reclassified without any new data.
- Prioritizing what to review: A case was presented in which ML selects, from the many CNVs called in WGS, only those an analyst actually needs to review. However, it performs best in regions with precedent in internal classification records and may miss novel CNVs without precedent.
- Prioritizing causal variants: A model trained on a large rare disease cohort ranked the causal variant at the top in most confirmed cases.
- AI that reads the literature (PEPPER): A score generated by an LLM reading publications (PEPPER) was combined with gnomAD data to produce a metric (OMELET) that complements LoF intolerance (constraint) for small genes that are hard to assess with existing metrics (LOEUF).
- Conversely, commercial LLMs alone fall short: When general-purpose LLMs such as ChatGPT were asked to classify variants, both accuracy and consistency were low, and they often mislabeled benign variants as pathogenic.
Why it matters
- VUS are among the most difficult results to manage in clinical practice. Data-calibrated AI is becoming a practical means of reducing VUS and speeding up interpretation.
- At the same time, patients and non-specialists increasingly use general-purpose AI to interpret test results on their own, making expert interpretation more important, not less.
What to expect in the clinic, and limitations
- As laboratories make greater use of AI, results will come back faster and ambiguous results will decrease.
- AI performance depends on the population makeup of its training data. Performance in Asian patients needs to be confirmed separately.
- When patients bring interpretations from general-purpose AI, these should be confirmed through a clinical laboratory’s report and genetic counseling.
5. Genomic screening: moving beyond diagnosis toward prevention
Genomic testing is expanding from “finding the cause in sick patients” to “finding risk in healthy people.” This year, the standards and operating models for doing so became much more concrete.
What was presented
- Standards for adult screening: Separately from its secondary findings (SF) list for diagnostic testing, ACMG is preparing a “Primary Findings (PF)” list for screening. ClinGen presented a framework for evaluating which gene–disease pairs should be included in screening.
- Real-world programs: At the opening plenary, three-year outcomes were presented from a US program of roughly 500,000 participants in which multiple health systems have offered CDC Tier 1 condition screening to research participants. From Europe, an integrated model was presented that spans pre-test education, counseling, prevention planning, and clinical referral.
- Newborn genomic screening: Results from large programs such as the Generation Study (UK) and GUARDIAN (US) were shared. Screen-positive rates differed by ancestry, and in East Asian newborns a hearing-related gene (GJB2) had a large effect.
Why it matters
- The central message of screening is that “genotype is not disease.” Because it is common for healthy adults to carry pathogenic variants, how results are communicated and managed matters as much as the test itself.
- As standards emerge where none existed, it will become easier for hospitals and countries to design screening programs.
What to expect in the clinic
- Programs that combine pre-test education, penetrance-aware result reporting, genetic counseling, and clinical referral pathways will become the standard.
- For newborn screening, panel design that reflects population-specific variant characteristics is essential. This is why more East Asian data, including from Korea, are needed.
Other notable updates
- Noninvasive fetal genome testing: By analyzing the entire fetal exome using only cfDNA from maternal blood, this approach reproduced most of the clinically significant variants found by invasive testing such as amniocentesis. It may offer a new option for families hesitant about invasive testing.
- Genetic causes of stillbirth: Many of the dominant causal variants found in stillbirth cases were inherited from healthy parents. Further validation is needed to determine whether these variants were the direct cause of stillbirth, but the findings call into question the assumption that “a variant inherited from a healthy parent must be harmless.”
- Diagnosis leading to therapy: There was active discussion of moving from ASO therapies designed for a single patient (N-of-1) toward broadly applicable ASOs that can treat many patients (N-of-many). The link between precise variant diagnosis and treatment opportunities is growing stronger.
- Falling sequencing costs: With a series of new sequencers coming to market, WGS costs are expected to keep falling, which may accelerate the use of WGS as a first-tier test.
Frequently asked questions (FAQ)
Q. Should a patient with negative WES/WGS go straight to long-read sequencing?
We recommend reanalysis of the existing data first, since new disease genes and analysis tools alone often provide the answer. If no answer is found and clinical suspicion remains high, a stepwise approach considering long-read sequencing, RNA analysis, or targeted testing that accounts for mosaicism is a practical next step.
Q. When is the right time for reanalysis?
Reanalysis should be done periodically, and promptly whenever the patient develops new symptoms or family history information changes. The more up to date the phenotype information, the more accurate the reanalysis.
Q. If an episignature test is negative, can the VUS be considered benign?
No. Episignatures can only be assessed for conditions with a known pattern, and many genes have no pattern. A detected signature is strong evidence for pathogenicity, but a negative result cannot be used as evidence for a benign classification.
Q. Can results be trusted when a laboratory uses AI to interpret variants?
In the laboratory, AI is validated against confirmed data and used within ACMG standards, and the final classification is made by experts together with other evidence. Used this way, AI can reduce VUS and improve the consistency of interpretation.
Q. What should I do if a patient brings an interpretation of their results from an AI such as ChatGPT?
Studies have shown that general-purpose AI alone has low accuracy and consistency in variant classification. In particular, it often labels harmless variants as pathogenic, which can cause unnecessary anxiety. We recommend confirming through the laboratory’s interpretation report and genetic counseling.
Q. Is it appropriate to recommend genomic screening to healthy adults?
Standards are being developed rapidly, but the system behind the test matters as much as the test itself. We recommend checking whether a program includes pre-test education, penetrance-aware explanation of results, genetic counseling, and clinical referral pathways, because many people who carry a pathogenic variant never develop the disease.
Q. Will newborn genomic screening replace conventional newborn screening?
For now, it is closer to a complement than a replacement. Its strength lies in detecting treatable conditions early that conventional tests struggle to catch, and panel design that reflects population-specific variant characteristics is essential.
Closing
These five trends point in one direction: patients getting faster and more definitive answers with fewer tests. 3billion will work to bring these changes into clinical practice quickly. If you have any questions or topics you would like to discuss in more detail, please feel free to contact us at any time.
Want to learn more about
3billion's genetic testing?
We'll reply within 1 business day.
