Bringing a Nearly Missed Diagnosis Back Into Focus – How 3ASC Redefined Automated Reanalysis

Product | 25. 12. 23

As time goes on, scientific literature and genetic databases continue to be updated.

Along the way, variants that were once classified as VUS are increasingly being reclassified as Pathogenic (P) or Likely Pathogenic (LP).
For this reason, regular reanalysis is becoming an ever more important part of genomic diagnostics.

Of course, for reanalysis to be truly effective, it is not enough for classifications to change.
What matters most is whether those updated variants are actually captured again based on the new knowledge and criteria.

Early automated reanalysis systems were largely designed around symptom similarity.
At the time, this was a reasonable and clinically sound approach.

However, as richer datasets and more advanced modeling techniques became available, it became clear that reanalysis systems themselves also needed to evolve.

In response, 3billion introduced 3ASC, an AI-based pathogenicity prediction model, as a new core criterion for automated reanalysis.
This allowed variants to be prioritized with much greater precision, ensuring that updated genetic knowledge could be reflected more quickly and meaningfully in real patient diagnoses.

The three real-world cases below show how this technical evolution led to clinically meaningful differences.

(Initial analysis: 2022–2023 / Automated reanalysis: October 2025)

Case 1. Even When a Variant Is Reclassified as Pathogenic, It Can Disappear If Symptoms Are Too Few

A six-month-old infant with seizures and a PNPO variant

Clinical context

A six-month-old girl presented with only one confirmed symptom: epilepsy.
At this early age, there was limited developmental or neurological information available.

Initial findings

One variant of uncertain significance (VUS) was identified in the PNPO gene.
At the time, it was considered insufficient to explain the clinical presentation, and the result was reported as Negative.

What changed over time

As additional studies and patient data accumulated, the variant was reclassified as Likely Pathogenic (LP).

The challenge

The patient lacked most of the characteristic features typically associated with PNPO deficiency.
If symptom similarity alone had been applied, this variant would likely not have met the threshold for reanalysis.

What 3ASC did differently

Rather than focusing on the number of symptoms, 3ASC prioritized the pathogenic signals inherent to the variant itself, elevating it to the top candidate.

Conclusion

👉 Even when a variant is reclassified as pathogenic, it may not be revisited if symptoms are too limited.

Thanks to 3ASC, this PNPO variant was re-evaluated during reanalysis and became a new clue for understanding the infant’s seizures.

Case 2.

Even When a Variant Is Reclassified as Pathogenic, It Can Be Missed Due to Nonspecific Symptoms
A child with developmental delay and a COQ7 variant

Clinical context

A 14-month-old girl presented with developmental delay and nonspecific MRI findings that were difficult to interpret clinically.
These were the only available clinical features.

Initial findings

Two variants were identified in the COQ7 gene, each inherited from one parent. At the time, both variants were classified as VUS.

What changed over time

As evidence accumulated, the two variants were reclassified as Pathogenic and Likely Pathogenic, respectively.

The problem that remained

Developmental delay and nonspecific MRI findings are extremely common in clinical practice.
If symptom similarity alone had been applied, this COQ7 variant would likely have been pushed down the priority list during automated reanalysis.

What 3ASC did differently

Regardless of how common the symptoms were, 3ASC focused on the pathogenic signals of the variant itself, and elevated this COQ7 variant to a high-priority candidate for reanalysis.

Conclusion

👉 Even when a variant is reclassified as pathogenic, common symptoms can still cause it to be deprioritized in automated reanalysis.

3ASC ensured that this COQ7 variant was not overlooked, ultimately providing an important clue to understanding the child’s developmental delay.

Case 3. Even When a Variant Is Reclassified as Pathogenic, It Can Be Excluded If It Does Not Explain Everything

A LOXHD1 variant that explained hearing loss—but not the full phenotype

Clinical context

A 13-year-old female patient presented with short stature, advanced bone age, and sensorineural hearing loss.
Although multiple symptoms were present, they did not initially point to a single diagnosis.

Initial findings

A VUS was identified in LOXHD1, a gene associated with hearing loss.
Because the remaining symptoms were unexplained, the case was reported as Negative.

What changed over time

The variant was later reclassified as Likely Pathogenic (LP), providing a clear explanation for the hearing loss.

The remaining issue

Short stature and advanced bone age are not typical features of LOXHD1-related disease.
When all symptoms were included in similarity scoring, the overall score decreased—putting the variant at risk of exclusion from automated reanalysis.

What 3ASC did differently

Instead of asking, “Does this explain all symptoms?”, 3ASC first asked, “Is there a variant that clearly explains at least one part of the phenotype?”

As a result, the variant remained in the reanalysis pool.

Conclusion

👉 Even when a variant is reclassified as pathogenic, it may still be overlooked if it does not explain the entire clinical picture.

Thanks to 3ASC, this LOXHD1 variant was not missed.
The diagnostic outcome moved from Negative to Inconclusive—not a final answer, but a meaningful step forward.

What these three cases have in common

Reclassification alone is not enough.
For a diagnosis to move forward, there must be a system that actively brings reclassified variants back into view.

3ASC ensures that clinically important pathogenic variants are not pushed out of the system—even when

symptoms are limited,
symptoms are common, or
not all symptoms align perfectly.

As clinical geneticists, we often experience moments when variants once considered uncertain gain new meaning as literature and databases are updated.
Automated reanalysis is a highly practical and clinically realistic tool that allows these changes to be reflected in real diagnostic decisions.

As more research and data continue to accumulate, meaningful clues will emerge for many more patients. And automated reanalysis will continue to play a critical role in that process.

Closing the Series – Rethinking Reanalysis

Across the five articles in this “Reanalysis at 3billion” series, we explored genomic testing from multiple perspectives, emphasizing one central idea:
genomic testing is not a one-time result, but an evolving process whose meaning expands over time.

In Part 1,
We addressed why unresolved cases and VUS are an unavoidable reality, and how the value of genetic testing grows as research, databases, and clinical guidelines continue to accumulate.
By reviewing the additional diagnostic yield reported in the literature, we aimed to answer a fundamental question: why reanalysis is not optional, but essential.

In Part 2,
We focused on how diagnoses can change not simply by rerunning data, but by incorporating new phenotype information.
We discussed which types of clinical updates truly matter for reanalysis, which tend to have little impact, and why a 2–3 year reanalysis interval is often the most efficient in practice.
This section addressed a very practical question:
when, what, and how to update clinical information.

In Part 3,
Through two illustrative cases, we showed how updated clinical information can expand interpretation toward a dual diagnosis.
At the same time, we explained that 3billion’s initial analysis is not a simple listing of variants, but a conservative and robust interpretation process that reports only what is clinically meaningful at the time.
This also helped clarify why NEG → POS conversions are relatively rare, and why it is crucial to distinguish between unreported symptoms and symptoms that are truly absent.

In Part 4,
We introduced the structure of an automated reanalysis system designed to continuously reflect knowledge updates across tens of thousands of samples and billions of variants.
We discussed how diagnostic potential can be managed strategically by categorizing different reanalysis scenarios, and why an AI-based pathogenicity prediction model (3ASC) was introduced to complement symptom-based approaches.
Ultimately, this section addressed why strategic, system-driven management is necessary—so clinicians do not need to manually track every update in an ever-expanding body of genetic knowledge.

Finally, in Part 5,
through three real-world cases, we demonstrated how AI-driven automated reanalysis can bring nearly missed variants back into focus—even when symptoms are limited, nonspecific, or not fully explained by a single gene.

From a clinical geneticist’s perspective, reanalysis is not a burdensome step of “rechecking old results,” but rather a practical and essential tool that connects a patient’s clinical course with the continuous evolution of medical knowledge, helping us move closer to more accurate answers.

We hope this series has offered clinicians who are considering reanalysis a small but concrete guide—
on when to update information, what to update, and how to leverage systems to maximize diagnostic benefit for their patients.Above all, we hope these five articles have conveyed one key message:
A single genetic test does not determine everything, and a patient’s diagnostic possibilities remain open—even as time passes.