Predictor scores

Predictor scores estimate one part of a variant's biological context. Some predict the possible effect of a missense change on a protein. Others measure how strongly a DNA position has been conserved across species. They do not diagnose a condition or calculate a person's chance of disease.

Gene Inspector Pro shows these scores as review signals. A higher or lower value can help decide what to inspect next, but it must be read with the variant consequence, population frequency, ClinVar submissions, inheritance pattern, genotype quality, functional evidence, and the relevant clinical context.

Available predictors

• REVEL: an ensemble score for rare missense variants
• AlphaMissense: a missense predictor using protein sequence, structure, and population data
• DANN: a genome-wide model that scores single-nucleotide variants
• SIFT: a missense predictor based on related protein sequences
• PolyPhen-2: a missense predictor using sequence and structural features
• GERP++: a conservation score for a genomic position
• PhyloP: a signed measure of evolutionary conservation or acceleration at a genomic position

Reading the colour bands

The green, amber, and red bands make a predictor's own scale easier to scan. They are not a shared scale of health risk. For example, a low SIFT score can suggest a protein change deserves review, while a high positive GERP++ score means the DNA position has been conserved. The signals are about different things and should not be averaged or treated as independent proof.

Practical limits

Predictors can disagree. Several scores reuse related conservation data or one another's outputs, so agreement does not always provide independent evidence. A low-concern prediction does not prove a variant is harmless, and a high-concern prediction does not establish clinical pathogenicity.