Is the ClinVar allele the odd one out at its own site?

At each of 120 ClinVar-selected coordinates in DNM1, PCCA, PCCB and MMUT, three substitutions are possible. For every Atlas scorer the pilot asks whether the clinically annotated one sits farther from a background profile than the other two, against a null that shuffles which substitution counts as the ClinVar one. Change the slice below and the test reruns in your browser.

120 sites × 3 substitutions = 360 variants22 scorers × 3 distance geometries = 66 comparisonschance that a given allele is farthest: 1/3

Genes

ClinVar allele's consequence

Null model

Shuffles

Multiple-testing family

Flagged site Drop chr9:128219079

Presets

Sites in this slice120

Comparisons passing BH q < 0.05–

Strongest comparison–

Where these numbers come from

Scorer × geometry

Each tile is one comparison. Colour shows how far the observed statistic sits from its own shuffle null, in null standard deviations. Select a tile to inspect it.

How often is the ClinVar allele the farthest?

Share of sites where the ClinVar substitution is the most distant of the three, with a 95% Wilson interval. Random labelling gives 1/3.

Context from the same cached outputs

What the rest of the audit reports

Coordinates alone already separate the groups

Gene labels barely cluster

Average-linkage clustering of per-site profiles scored against the four gene labels by adjusted Rand index (ARI), with label-shuffle p-values.

Training-data checks, as reported