1. Introduction
Variant classification is prone to cognitive biases and logical fallacies. Even experienced curators can fall into traps that lead to misclassification. In this post, we dissect the five most common errors and how to avoid them.
2. Double Counting (Circular Evidence)
This is the cardinal sin of variant interpretation: using the same piece of data to satisfy multiple criteria.
Using a functional assay to apply PS3 (Functional Studies) AND PP3 (In-Silico) because the in-silico tool was trained on that functional data.
Apply PS3 only. The functional data supersedes the prediction.
3. Ancestry Bias (The Manrai Artifact)
Assuming that absence in a database means the variant is rare in all populations.
The Trap
You find a variant absent in gnomAD (which is 50% European). You apply PM2. However, your patient is of African ancestry, and the variant is actually common (5%) in that specific sub-population, which is underrepresented in gnomAD.
Solution: Always check the specific ancestry population in gnomAD and consider All of Us data.
4. In-Silico Overreliance
"It's red in PolyPhen, so it must be pathogenic."
Reality Check: Most in-silico tools have a specificity of ~60-70%. They are screening tools, not diagnostic tools. Never use PP3 as the sole evidence to upgrade a VUS to Likely Pathogenic.
5. PVS1 Misuse
Applying PVS1 to any truncation without checking the NMD rules.
- Last Exon: Truncations in the last exon usually escape NMD. The protein is made, just shorter. Is the missing tail critical? If not, PVS1 does not apply.
- Splice Rescue: A splice site mutation might just cause exon skipping that leaves the reading frame intact (e.g., skipping an in-frame exon).
6. The Phenotype Trap
Confirmation bias is powerful. If a patient has Marfan syndrome and you find a VUS in FBN1, your brain wants to call it Pathogenic.
The Discipline: You must evaluate the variant blind to the patient's phenotype first. The phenotype only comes in via PP4 (Phenotype specificity), which is a weak criterion.