Twelve preprints, seven of the first ten negative results.

Folding stability prediction misses three of four known loss of function variants in the SCN5A N terminal domain

The method I had been relying on only works when the cause is a buried charge. Given four variants already known to break this channel, it caught one.

Closed the folding corrector route, which is the one I most wanted to work.

Editability scored off target counting, and the specificity cost of relaxing the docking rule

Counting the places a base editor might act by mistake, by asking whether it could actually edit there rather than whether the sequence merely looks similar, removes most of the apparent risk. It also shows the naive method is biased toward recommending the worse of two editors.

Underpins the base editing route. The method is constructive; the finding that no guide here is clean is not.

Deposited functional evidence in ClinVar is searchable only by exact term, and no surveyed tool reads it

Laboratories deposit real measurements of what variants do into the public database. Most of the tools that interpret variants already have the full record on disk and never look at the field. The evidence is public and effectively invisible.

The one result on this site that has nothing to do with my variant. The census page sets out its size.

No antisense oligonucleotide has demonstrated target engagement in a human cardiomyocyte

Every design I tested for editing the message rather than the gene either changed neighbouring letters it should not have, or bound in too many other places. Separately, no molecule of this class has been shown to reach a human heart muscle cell at all.

Sets the ceiling on the RNA editing route.

The Conflicting classification of SCN5A p.Arg104Gln rests on one outdated submission, not on divided evidence

My variant is labelled Conflicting in the public database, which reads as genuine disagreement between laboratories. It is not. Across every available assay it behaves the same as its neighbour, which is classified pathogenic.

An argument that the evidence should be reviewed. Only the submitting laboratories and the expert panel can change a classification.

SCN5A lacks the non productive mRNA reserve that antisense upregulation therapy would need in human heart

A therapy that boosts output from the healthy copy needs a reserve of wasted message to redirect. Across 827 heart samples that reserve is about 300 times smaller than in the brain, where the approach already works in children.

Closed the upregulation route. It killed one method, not a level.

Peak current alone cannot separate trafficking arrest from coupled gating in SCN5A dominant negative variants

The standard assay cannot tell apart two different ways a broken copy interferes with a healthy one, because the two explanations are algebraically degenerate: they predict the same number. Measuring harder with the same method cannot help.

Names the experiment that would settle the open question, which is costed in full.

Neither local strain nor exposed hydrophobic surface explains pathogenicity at SCN5A Arg104

Two candidate explanations for why this position matters were tested and eliminated. What survives is that this residue anchors a buried acidic pocket unchanged across species. The paper also corrects several of my own earlier claims about the geometry.

Eliminates two explanations for the mechanism, and retires one I had published myself.

A stability predictor with known blind spots nominates seventeen uncertain SCN5A N terminal variants for testing

Of 131 variants of uncertain significance in this region, 17 score above a measured noise floor and are worth testing. The other 114 are uninformative, which is not the same as benign, and that matters to anyone carrying one.

A usable shortlist, built on top of the measured blind spot that makes a negative score meaningless.

A large scale functional scan of this gene does exist, and has been under construction for about nine years. It now covers roughly 45 percent of the protein, concentrated on the pore-forming regions of the four transmembrane domains, and its author confirms that it does not cover the domain this shortlist addresses (personal communication, 13 August 2026). So the region here is not merely unmeasured. It is the region the field's largest functional effort has left uncovered, because that effort went to the pore. The blind spot below is unchanged by this, and the 114 uninformative scores still mean nothing at all.

Residue centroid distance misses close atom contacts in the SCN5A N terminal domain

Models that measure distance between the centres of residues miss about 9 percent of real atomic contacts, and disproportionately the long side chain and salt bridge ones.

A methods correction to a published penetrance model, which cannot see the contact at my own position at all.