Every variant-interpretation tool on the market today draws from the same well: human population frequency, clinical case reports, and computational predictions trained on those same two things. When all three agree, a variant is usually easy to call. When they don't (or when there isn't enough of any of them) the result is a growing pile of variants of uncertain significance. Orthogonal evidence is a different kind of answer to that problem: not a better model built on the same data, but an independent measurement from a completely different source.
What "orthogonal" actually means here
In statistics and engineering, orthogonal means independent... two axes that don't move together, so a signal on one tells you something a signal on the other couldn't. Applied to variant interpretation, orthogonal evidence is evidence gathered by a method that's conceptually and technically independent of population-frequency databases or conservation-based scores. It isn't a new weighting of the same signal. It's a separate observation you layer on top of ClinVar, gnomAD, or an ensemble predictor like REVEL, not a replacement for any of them.
For CodeXome, that orthogonal layer is direct evolutionary observation across primates: did this exact amino-acid substitution occur, and persist, in other primate lineages across the roughly 87 million years since our common ancestor with the wider primate order? If it did, that's strong evidence biology has already tolerated it. If a position never varies across any of the 55 genera in that record, that's a different kind of evidence entirely, a site biology has never let change. (Why the window has to be primate-wide rather than human-to-human is its own question — see Why Human-to-Human Comparisons Miss 99.8% of Evolutionary History.)
Isn't this just another conservation score?
It's the question we get most often, and it deserves a direct answer: no. A conservation score is a statistical summary. How invariant a position looks across an alignment, averaged into a single number. It can tell you a site is "probably important," but it can't tell you which specific changes, if any, were actually survived by a living lineage.
Orthogonal evidence, in this sense, is a recurrence record, not an average. It shows the exact substitution, in the exact residue, observed and persisting in a real population of a real species, not an inference about how conserved a region looks in general. That distinction matters most exactly where a conservation score is least useful: sites that look moderately conserved on paper, where a specific human variant either has or hasn't already been tried by nature.
What it looks like in practice
In an internal validation study run against the GIAB (Genome in a Bottle) reference dataset, cross-referencing variant calls against CodeXome's primate database removed roughly 75% of natural variation in a single filtering step, variation shared with other primate lineages, and therefore already tolerated by biology. What's left is a shorter list, and that's where the real work of variant curation actually needs to happen.
That filtering step is only as useful as the data behind it. Orthogonal evidence is only as credible as its provenance: the primate sequences need to come from a single, controlled pipeline, mapped consistently to the same reference coordinates, so "this substitution occurred in this species" is a comparable claim across the whole dataset. An orthogonal signal built on noisy or inconsistent data isn't independent evidence, it's just a second source of noise.
Where the evidence agrees, and where it doesn't
Used well, orthogonal evidence does two things. When it agrees with existing calls, it strengthens confidence. A variant already flagged benign by population data that also turns out to recur across primate lineages is now supported by two independent lines of evidence instead of one. When it disagrees, it's worth a second look either way: a variant with weak population support that turns out to be tolerated across the primate order becomes a strong candidate for reclassification toward benign; a variant at a site that has never varied across 87 million years, despite ample opportunity, becomes a stronger candidate for functional follow-up instead.
This is the same principle behind CodeXome's published reclassification numbers. Roughly 20% of ClinVar VUS reclassified toward likely benign on average across the genome, and closer to 30% within BRCA1 and BRCA2 specifically, as detailed in The VUS Bottleneck Explained. Each one a case where the orthogonal signal gave a confident answer where population and clinical data alone had left a gap.
What this is, and isn't
Orthogonal evidence is not a clinical verdict, and CodeXome isn't a diagnostic. Nothing here replaces the ACMG/AMP framework a lab is required to follow for a clinical call. What it is: research-grade, actionable evidence that informs prioritization and hypothesis generation. A way to spend less time on the large fraction of variants biology has already tested, and more time on the smaller number that actually need a closer look. (For the fuller case on why this counts as a genuinely independent evidence axis, see Why Evolution Is the Missing Dimension in Variant Interpretation.)
Think of it as a second opinion from nature itself: not the final word, but an independent check on the calls you're already making, built from watching what happened for 87 million years before anyone needed to ask.
If you want to see how this plays out on your own data, evaluate CodeXome directly.