Skip to main content
In Development
🚀 Coming Soon! Seeking Testers

We're seeking organizational testers for early access to our AI-powered genomics platform.

Part 4 of 6
Variant Classification Series

The Bayesian Framework: Moving Beyond Rules to Probabilities

November 23, 2025
18 min read
RW

Ryan Wentzel

Founder & CEO, Humanome.AI

1. Introduction

The 2015 ACMG/AMP guidelines provided a necessary common language, but their "rule-based" logic (e.g., "1 Strong + 2 Moderate") has significant limitations. It treats evidence as categorical rather than continuous, leading to "cliff-edge" effects where a variant falls just short of a classification despite overwhelming evidence.

2. Limitations of the 2015 Rules

Consider a scenario where you have 6 pieces of "Supporting" evidence. Under the 2015 rules, this might not sum to "Pathogenic" because the rules explicitly ask for Strong or Moderate evidence. Yet, statistically, 6 independent lines of supporting evidence yield a posterior probability of pathogenicity >99%.

This rigidity led the ClinGen Sequence Variant Interpretation (SVI) working group to develop a more flexible, mathematically grounded approach.

3. The Tavtigian Framework (2018)

In 2018, Tavtigian et al. published a Bayesian framework that calibrated the ACMG criteria. They assigned an "Odds of Pathogenicity" (OddsPath) to each strength level:

  • Very Strong: OddsPath = 350:1
  • Strong: OddsPath = 18.7:1
  • Moderate: OddsPath = 4.33:1
  • Supporting: OddsPath = 2.08:1

By assuming a prior probability of 0.1 (10%), one can calculate the posterior probability using these odds.

4. The Point System

To make this usable in a clinical lab without complex calculators, the OddsPath values were converted into a simple additive point system.

ACMG Points

ClinGen SVI Compatible
PVS1 (Very Strong)+8
PS (Strong)+4
PM (Moderate)+2
PP (Supporting)+1

Benign Points (Negative)

BA1 (Stand-alone)-8
BS (Strong)-4

5. Classification Thresholds

Once you sum the points, the classification is determined by these thresholds:

ClassificationPoint RangePosterior Probability
Pathogenic≥ 10> 99%
Likely Pathogenic6 - 990% - 99%
VUS0 - 510% - 90%
Likely Benign-1 to -6< 10%
Benign≤ -7< 0.1%

6. Resolving Conflicts

What happens if you have PVS1 (+8 points) and BS1 (-4 points)?

The Conflict Rule

If a variant has both Pathogenic and Benign criteria applied, it defaults to VUS unless the conflict can be resolved.

Example: A frameshift variant (PVS1) is found at 1% frequency in gnomAD (BS1).
Resolution: The high frequency suggests the frameshift is not disease-causing (perhaps it's in a pseudogene or an exon not expressed in the relevant tissue). The PVS1 should likely be removed or the variant classified as Benign.

7. Case Study: BRCA1 c.68_69del

Let's apply the point system to the famous 185delAG founder mutation.

  • PVS1 (+8): Frameshift leading to NMD.
  • PS4 (+4): Significantly enriched in breast cancer cases vs controls.
  • PP1 (+1): Cosegregates with disease in multiple families.

Total Score: 8 + 4 + 1 = 13 Points

Result: Pathogenic (>99%)

Tags

Bayesian Statistics
ClinGen
SVI Framework
Scoring
Humanome.AI - Genomic Variant Intelligence Assistant