The Clinical Geneticist’s “Third Eye”: How AI-Driven Automated Reanalysis Unlocks New Diagnostic Possibilities

In the previous article, we explored how diagnostic insights can expand over time—for example, a patient initially suspected of having NF1 was later found to have a dual diagnosis (NF1 + FANCA), and another case involving a MOCS1 variant was reinterpreted thanks to updated biochemical knowledge. These examples highlight a key message: up-to-date clinical information is one of the strongest drivers of successful reanalysis.

At the same time, we also discussed the practical challenges. Requesting reanalysis too frequently—with no meaningful change in the patient’s clinical condition—can be inefficient. And it is simply unrealistic to expect every clinician to continuously keep pace with the rapidly evolving landscape of genomic science.

So the conclusion becomes clear.
The most practical and powerful way to maximize diagnostic yield is strategic management supported by a robust system.

Today’s article focuses on that very system—the core of 3billion’s approach. We will take a closer look at how 3billion’s Automated Reanalysis works in real clinical practice, and what long-term benefits it offers to patients.

---

The Never-Ending Evolution of Knowledge: 3billion’s Automated Reanalysis

Reanalysis is the process of revisiting previously analyzed genomic data using the most up-to-date medical and genetic knowledge—new disease and gene discoveries, updated clinical databases, emerging scientific literature, improved algorithms, and more.

At 3billion, this process is not a one-time effort but a continuously running, automated system that supports patients throughout their entire diagnostic journey until a final diagnosis is reached.

• Total accumulated samples: ~75,000 (as of March 2025)
• Undiagnosed (Negative/Inconclusive) samples: ~45,000–52,500
• Total variants under reanalysis: 3.38 to 3.93 billion

Managing and extracting meaningful diagnoses from this astronomical volume of variants requires a highly structured and strategic approach.

---

The Core of Automated Reanalysis: 9 Strategic Classification Scenarios

At 3billion, reanalysis is never performed randomly. To ensure that clinically meaningful variants are not overlooked, we categorize situations requiring reanalysis into 9 major scenarios and 26 detailed subcategories.
(These classifications are continuously updated as scientific knowledge evolves.)

Examples of major categories include:

• Variants that have been reclassified as Pathogenic or Likely Pathogenic
• Variants in genes newly added to OMIM that match the patient’s phenotype
• High-impact VUS that closely fit the clinical presentation
• Variants in genes that are clinically suspected based on phenotype

Thanks to these granular and strategic criteria, our system can selectively re-examine only the variants that truly matter—ensuring efficient yet highly targeted reanalysis within an enormous dataset.

---

Key System Upgrade: Overcoming the Limits of Symptom Similarity with 3ASC

In the early stages of our automated reanalysis system, variant detection relied heavily on symptom similarity scores. Variants were flagged when the patient’s clinical features showed sufficient similarity to known disease phenotypes.

However, this approach came with inherent limitations:

• When a patient had only a few symptoms—or symptoms that were nonspecific—the similarity score could be artificially low.
• In dual diagnoses, a single variant often cannot explain the full clinical picture, leading the system to overlook important findings.
• Ultimately, this meant that clinically significant variants could be missed.

The Solution: Introducing AI-powered 3ASC

Through extensive analysis of these “missed cases,” 3billion discovered a crucial pattern:
the overlooked variants consistently ranked within the Top 5 of the 3ASC algorithm.

Based on this insight, we updated our criteria accordingly.

---

What Is 3ASC?

3ASC is an AI-driven pathogenicity prediction model developed by 3billion.
Trained on hundreds of thousands of real-world cases, it provides a highly refined estimate of how likely a variant is to cause disease.

Unlike traditional interpretation methods—which often rely mainly on features such as whether the variant lies in a functionally important domain or whether an amino acid is evolutionarily conserved—3ASC integrates a broad range of clinical and genomic features, including:

• ACMG Bayesian score
• Variant Allele Frequency (VAF)
• 3billion’s in-house population frequency
• Gene-level intolerance metrics (pLI, LOEUF)
• Computational pathogenicity predictions
• ClinVar annotations
• And numerous other specialized genomic features
  

By synthesizing all these elements, the AI model classifies each variant into High, Mid, or Low tiers and highlights those most likely to be clinically actionable.

---

Updated Reanalysis Criteria

• Previous approach: Variants were selected based on meeting a symptom similarity threshold.
• Improved approach: Variants with a 3ASC score of Mid or higher are now included—capturing clinically important variants that symptom-based methods alone may miss.

Results: Higher Accuracy, Broader Diagnoses, Fewer Misses

Since the introduction of 3ASC, the performance of our automated reanalysis system has shown clear and measurable improvement:

1. 5% increase in automated reanalysis accuracy

2. 10 out of 24 reports issued after the 3ASC upgrade were detected because of 3ASC

3. Strong impact especially in difficult-to-diagnose cases, such as:

• Infants with very limited symptoms at the time of initial testing
• Potential dual-diagnosis cases
• Patients with incomplete or sparse phenotype information

In short, 3ASC has become a powerful safeguard—ensuring that clinically meaningful variants are not missed, even when symptom information is imperfect.

---

Automated reanalysis reduces the clinical burden and maximizes diagnostic yield

Genomic knowledge evolves continuously. Yet it is unrealistic for clinicians to track every update in real time.
3billion’s automated reanalysis system bridges this gap by integrating both clinical updates and genomic knowledge updates, allowing diagnostic potential to grow over time rather than stagnate.

With 3ASC now embedded in the system, we can provide something truly meaningful:
a long-term diagnostic care model in which a patient’s likelihood of receiving a diagnosis increases as time passes.

---

In the next article, we will introduce real-world examples of how this AI-powered automated reanalysis system has provided diagnoses to previously unsolved cases.

Curious about reanalysis?

Explore more insights from 3billion’s clinical geneticists.

1. 3billion Reanalysis Part 1: Genetic Testing Is Not the End — Why Is Reanalysis Necessary?
2. 3billion Reanalysis Part 2 : New Clinical Information Can Change the Diagnostic Direction
3. 3billion Re-analysis Part 3: Clinical Updates Expand Diagnosis – Case Sharing