Worked Coding-Locus Annotation and Evidence Integration

From a Variant List to a Locus Interpretation

A coding annotation can look decisive: a variant is labelled missense, a score is high, and the locus is associated with a phenotype. Yet these observations answer different questions. Allele harmonization asks whether records refer to the same alternate allele. Transcript annotation asks what that allele changes in a particular transcript. A functional score estimates an outcome defined by its model or calibration data. Fine-mapping weights describe uncertainty about which candidate variant is causal under the association model. None of these alone establishes that a variant causes the phenotype through a particular molecular mechanism.

A careful locus interpretation therefore preserves the distinctions among identity, consequence, score, and association evidence. The examples below use entirely hypothetical variants and measurements. They illustrate a coding-locus workflow, not a claim about a real gene or study.

Step 1: Harmonize the Variant and Effect Allele

Begin by confirming the genome assembly and variant representation. For a single-nucleotide variant, record chromosome, position, reference allele, and alternate allele on the stated reference strand. Then align the association effect allele to the same orientation. A reported effect estimate applies to the effect allele relative to the comparison allele; if the effect allele is accidentally assigned to the wrong strand or allele, both the direction and the biological interpretation can be wrong.

For substitutions, reverse-complementing changes each base to its complement and reverses the allele representation. For example, genomic T>C corresponds to A>G when represented on the opposite strand. Palindromic substitutions (A/T and C/G) require particular care: their reverse complements have the same unordered allele pair, so strand cannot be resolved from the pair alone. Reliable build, strand, or frequency information may be needed. Indels also require a consistent representation, including normalization against the same reference sequence.

Step 2: Resolve Consequences in Transcript Context

A genomic variant does not have one universal coding consequence. Consequence depends on which transcript is considered, its strand, its exon structure, and the coding sequence used. A substitution can alter a protein in one transcript but fall outside the coding sequence of another. Even when two transcripts are both coding, they may use different exons or reading frames and yield different amino-acid changes.

For each candidate variant, record the transcript identifier and version, whether the variant overlaps its coding sequence, the coding-DNA change, and the predicted protein consequence where applicable. State which transcript is being used and why it is relevant to the question—for example, whether it is expressed in a relevant tissue or is a specified reference transcript. A label such as “missense” is incomplete without that context. It also describes a sequence consequence, not demonstrated alteration of protein activity or trait causation.

Step 3: Compare Functional Scores on Their Defined Scale

Functional scores are interpretable only with their scale and scope. A percentile rank says how a variant compares with a reference distribution; it is not a probability. A calibrated probability refers to a particular target outcome and calibration population. For example, a score calibrated against an assay-defined damaging label estimates that label under the calibration setup, not the probability of causing a human trait.

Check that the score applies to the variant class and transcript consequence being interpreted. Also note its reference set, calibration target, and direction (whether larger values indicate stronger predicted impact). Two scores can agree because they capture overlapping sequence features. Agreement can strengthen a descriptive summary, but it does not justify treating them as independent evidence and multiplying their values or odds.

Why Posterior Weighting Works—and What It Means

Let \(C\) denote the causal variant under the fine-mapping model, let \(w_i=P(C=i\mid D)\) be the posterior probability assigned to candidate variant \(i\) given association data \(D\), and let \(a_i\) be a numerical annotation or score for that variant. If the weights cover mutually exclusive candidates and sum to one, the posterior expectation of the annotation of the causal variant follows from conditioning on which candidate is causal:

The second equality holds because, when \(C=i\), the annotation of the causal variant is \(a_i\). This justifies a posterior-weighted summary as an expected annotation value under the stated model. It does not convert an annotation into evidence that independently updates the causal posterior. Nor does it account automatically for uncertainty in the annotation itself, transcript choice, score calibration, or the fine-mapping assumptions.

If posterior weights cover only a credible set with total mass below one, report the total weight and the conditioning explicitly. If some scores are unavailable or inapplicable, do not treat those entries as zero: either report the unscored posterior mass and a mean conditional on score availability, or use a clearly justified alternative summary. A missing score is not evidence of no functional effect.

Step 4: Integrate Evidence Without Overstating It

A locus-level account should put the pieces in a traceable order. First establish that the association and annotation records identify the same allele. Next report consequences by transcript, including competing transcript interpretations. Then summarize score values on their actual scale and say which variants are eligible. Finally, place these observations beside the fine-mapping weights and state the residual uncertainty.

The evidence may converge without proving a mechanism. A variant can have high posterior weight and a severe predicted consequence, but a score remains a prediction about its defined target. Conversely, a less severe label does not exclude functional relevance through an unmeasured effect or transcript. Molecular assays, relevant tissue context, and perturbation evidence may change the assessment, but they should be described as their own evidence rather than inferred from a coding label.

Common Pitfall: Treating a Severe Label as a Causal Conclusion

Wrong reasoning: “This is the highest-scoring missense variant, so it is the causal variant and the gene acts through protein damage.” The conclusion skips several uncertainties: whether the allele was harmonized correctly, whether the chosen transcript is relevant, what the score was calibrated to predict, and how much posterior probability the variant carries relative to alternatives.

Correction: report the allele alignment, transcript-specific consequence, applicable score and scale, and fine-mapping weight separately. Then state what their combination supports—for example, that the variant is a prioritized candidate with a high predicted assay-defined impact score—while retaining uncertainty about causality and mechanism. Functional severity is evidence to interpret, not a substitute for causal evidence.

A useful final summary is compact but explicit: identify the candidate allele and representation; name the transcript and predicted coding change; give the score with its meaning; report the posterior weight or weighted annotation summary; and mention alternatives or missing evidence. Such a summary lets another analyst distinguish a robust allele-level result from a conclusion that depends on transcript selection or score assumptions.