A genetic report may describe a variant as “pathogenic,” “likely pathogenic” or a “variant of uncertain significance.” Those words answer a question about the evidence for that variant. Understanding what the result means for a particular person takes another step: connecting it with the condition, inheritance pattern and clinical findings.
This guide explains how to read that evidence without trying to classify a personal result from a checklist. It focuses on inherited, primarily single-gene conditions. Cancer treatment biomarkers, large copy-number changes and other types of findings may require different interpretation frameworks.
What the five categories mean
The ACMG/AMP framework gives laboratories a common vocabulary. Here is a plain-language reading of its five categories. Richards and colleagues’ consensus guidance explains their intended clinical use.
| Category | How to read it |
|---|---|
| Pathogenic | The evidence supports a disease-causing role in the stated context. |
| Likely pathogenic | The evidence strongly favors a disease-causing role, with some uncertainty remaining. |
| Uncertain significance (VUS) | The available evidence does not establish whether the variant is disease-causing or benign. |
| Likely benign | The evidence favors a non-disease-causing interpretation. |
| Benign | The evidence supports a non-disease-causing interpretation. |
A VUS is not a positive diagnosis and should not itself determine clinical decisions. A clinician may still recommend care because of symptoms or family history; that is a separate justification from treating the uncertain variant as pathogenic.
Read the report in three passes
- Identify the finding. Copy the gene, exact variant description, reference transcript and version, and the laboratory’s classification date. Keep the full report, not just a screenshot of one row.
- Identify the question. Which condition is being assessed? Is the laboratory discussing an inherited disorder, carrier status or another finding? Write down what you understand the result to answer.
- Identify the next conversation. What remains unresolved, who can clarify it, and what additional information would be useful? Bring those questions to the reporting laboratory or genetics professional.
Our variant research worksheet helps organize search terms and evidence questions. Use public variant descriptions only; do not enter identifying patient information.
Why the criteria need gene-specific judgment
The original 2015 framework is the foundation, not a complete set of instructions for every gene. ClinGen expert groups refine criteria for particular gene–disease relationships. Appropriate population-frequency limits, functional assays and interpretation of family observations can differ between conditions. The overview of ClinGen specifications explains why those adjustments matter.
Before applying a rule, check the ClinGen Criteria Specification Registry. Confirm the gene, condition, inheritance pattern, release status and version. A specification still in development should not silently be treated as a released recommendation. Record the version you used so another reviewer can reproduce the reasoning.
Example 1: rare does not automatically mean harmful
Fictional teaching example. A student finds a missense variant—a change that substitutes one amino acid—in a gene related to the condition being studied. It is absent from a population database, and several prediction tools label it damaging. The student writes, “This must be the explanation.”
Two questions are missing. How much weight does absence from that database deserve? And are the predictions genuinely independent pieces of evidence?
ClinGen’s general recommendation reduced the absence/rarity criterion PM2 from moderate to supporting strength. Rarity by itself can be overvalued; the applicable gene-specific rules and population-data limitations still need checking. See the ClinGen PM2 recommendation.
Several software outputs may rely on overlapping information. Counting each as a separate experiment can exaggerate the evidence. The Bayesian analysis by Tavtigian and colleagues helps explain this problem: combining evidence assumes an appropriate relationship between the observations, rather than treating every label as a new vote.
A better note would be: “The variant is a candidate. I need to document the population evidence, the prediction method and the missing clinical or experimental evidence before reaching a classification.” The example does not assign a formal category to a real variant.
Example 2: a stop signal still needs context
Fictional teaching example. Two reports describe variants predicted to introduce an early stop signal. One affects a well-established disease-relevant transcript; the other affects an exon whose relevance is uncertain. It would be tempting to apply the same evidence strength to both because their annotation says “stop gained.”
The PVS1 criterion concerns loss of function where that is an established disease mechanism. Its strength depends on matters such as the relevant transcript, the position of the stop signal, possible RNA degradation and whether a functional product could remain. The ClinGen PVS1 recommendations provide a decision process rather than a blanket rule for every truncating change.
For the comparison, draw three columns: “what the annotation predicts,” “what the disease mechanism requires” and “what we still need to establish.” That small separation prevents a description of a sequence change from becoming an unsupported clinical conclusion.
Example 3: an abnormal experiment is not the whole answer
Fictional teaching example. A paper reports reduced protein activity for a candidate variant in cultured cells. Before calling that strong evidence, ask whether the experiment measures the disease-relevant function, whether suitable controls were included and whether its performance has been established.
ClinGen’s PS3/BS3 functional-evidence recommendations distinguish an interesting laboratory observation from an assay that can support a clinical classification at a justified strength. A normal result also needs scrutiny: an assay may not test every mechanism through which a variant could act.
RNA splicing evidence has additional guidance. The ClinGen splicing recommendations explain how observed transcript changes can inform PVS1 and why the same splice effect should not simply be counted again as separate functional evidence. Our clinical RNA-sequencing guide develops that example.
A worksheet for a journal club or laboratory discussion
Use one row per observation. Writing the uncertainty beside the evidence makes it easier to see what a new experiment would actually resolve.
| Write down | A useful question |
|---|---|
| Variant, transcript and condition | Are all sources discussing the same finding and disease relationship? |
| Source, date and guideline version | Can another reader locate the evidence and reproduce the rule choice? |
| Observation and proposed criterion | What was actually measured, and why does that criterion apply? |
| Evidence strength and limitations | Which assumption would most change the weight assigned? |
| Overlap with other evidence | Have we counted the same event, family or prediction more than once? |
| Unresolved information | What specific evidence could change the conclusion? |
When two studies appear to disagree, use the paper-comparison workspace to separate their methods and outcomes. When laboratories disagree, our ClinVar guide explains how to inspect the individual submissions.
What to ask after receiving a report
- Which part of the result explains the reason I was tested, and which part remains uncertain?
- What evidence supported the classification?
- Would additional clinical or family information help the laboratory?
- What is the laboratory’s policy for reassessment and notifying me of a change?
You do not need to master every evidence code to ask those questions. A useful explanation should let you distinguish the laboratory’s conclusion, the evidence behind it and the clinical decisions that still need discussion.
Worked example: what a Bayesian update actually does
A Bayesian model asks how new evidence changes a starting estimate. Tavtigian and colleagues’ 2018 analysis expressed ACMG/AMP evidence strengths in this framework. The arithmetic can clarify the reasoning, but it cannot decide whether an assay is valid, whether a gene causes the condition, or whether two observations are independent.
The following numbers are invented for teaching. They are not a real variant assessment or a mapping of “strong” or “moderate” ACMG evidence. Assume a starting probability of 10%, solely to make the calculation easy. In odds form, that is 1 to 9. Now suppose a genuinely informative result has a likelihood ratio of 9: it is nine times as likely under the pathogenic explanation as under the benign explanation.
- Convert probability to odds: 0.10 ÷ 0.90 = 1/9.
- Apply the assumed evidence: (1/9) × 9 = 1.
- Convert the updated odds to probability: 1 ÷ (1 + 1) = 50%.
If a second result is appropriately independent and also has a likelihood ratio of 9, the odds become 9 to 1, or 90%. Recounting the first experiment in a second paper does not supply a second independent result. Nor can three prediction websites be assumed to represent three independent experiments.
The starting estimate matters too. With an assumed prior probability of 1%, the same likelihood ratio of 9 gives 9/99 odds, or about 8.3% probability. Neither starting value should be selected merely to achieve a desired classification. This toy calculation illustrates why calibration and scope must accompany the number.
How the points model relates to the arithmetic
The 2020 points-model paper uses points proportional to log odds: supporting, moderate, strong and very strong pathogenic evidence correspond to 1, 2, 4 and 8 points. Supporting and strong benign evidence contribute −1 and −4. Its proposed categories place 6–9 points in likely pathogenic and 10 or more in pathogenic, under its stated assumptions. This is a formal model, not permission to assign evidence strengths from a web search.
Hypothetical evidence meeting: a reviewer proposes one strong and one moderate pathogenic criterion. Their arithmetic is 4 + 2 = 6. The next reviewer should inspect why each criterion is justified, whether they describe independent evidence, whether contrary evidence exists, and whether the applicable gene-specific specification permits the combination. Correct addition cannot rescue an invalid criterion. The original 2015 combining rules and later quantitative models are not identical in every combination; document which framework is being used.
Do not mix different kinds of points. The ClinGen de novo recommendation uses an internal scoring process to choose the strength of PS2/PM6. It explicitly states that those observation points are not equivalent to the Tavtigian classification points. Convert evidence through the applicable recommendation, rather than adding two incompatible scoring systems.
What the probability does—and does not—describe
A modeled probability that a variant is pathogenic is not the probability that a particular person will develop disease, transmit it, respond to treatment or have a severe outcome. Those questions need inheritance, penetrance and clinical context. Our penetrance and expressivity guide explains this additional step.
When a laboratory meeting ends in disagreement, record the exact variant and transcript; the gene–disease relationship; the released specification and version; the underlying observations; proposed codes and strengths; any overlap; and the unresolved point. For example: “We agree on the assay result but disagree about whether validation supports moderate or strong evidence.” That is more useful than recording only two competing totals.
The ClinGen Criteria Specification Registry is the place to check applicable released specifications. Use our variant research worksheet to organize the discussion, and the paper-comparison worksheet when evidence comes from studies with different methods. The purpose is a reproducible explanation of uncertainty, not a consumer calculator for changing a laboratory result.