Blindspot Genetics — Evidence Base

The figures on the home page

All four are measured from the Blindspot database, not taken from a publication. They were last measured on 2 October 2026; the home page reads the stored values and shows the date.

1 in 3

6,337 of the 19,250 genes Blindspot carries have at least one of five blind spots the platform measures:

coding sequence standard short reads can’t reliably call (class 3–5): 2,181
a documented pathogenic deletion or duplication in ClinVar, or ClinGen dosage sensitivity: 3,286
a coding exon lying entirely in repeated sequence: 2,263
a promoter in GIAB’s hard-to-sequence promoter set (BadPromoters): 962
a promoter less than half benchmark-checkable: 743

Many genes have more than one, so the rows overlap. 6,337 is 32.9%, so “1 in 3” rounds up. The kinds are not equally severe: a copy-number gene is a blind spot for small-variant calling, not necessarily for a laboratory that runs copy-number analysis on the same reads. ClinVar deletion reports also cluster in well-studied disease genes; on the 5,694-gene clinical panel the share is about half.

ClinVar · ClinGen · GIAB v3.6 · Blindspot database

1 in 9

2,181 of the 19,250 genes Blindspot carries are Reference-sensitive, Technology-sensitive or Refractory: part of their coding sequence cannot be reliably called with standard short reads on GRCh38, on T2T-CHM13, or on both. Each class points to a different fix — another reference genome, long-read sequencing, or a separate assay. This figure covers coding-sequence callability only; copy-number and promoter blind spots are measured separately. The classes are defined on the definitions page.

Blindspot database · release v1.9

1 in 5

66,158 of 308,692 ClinVar pathogenic and likely-pathogenic variants on GRCh38 lie in a GIAB difficult region — segmental duplication, low mappability, extreme GC, tandem repeat or homopolymer. That is 21.4%, so “1 in 5” rounds down. It counts variants that sit in hard sequence; it does not say they were missed.

NCBI ClinVar · GIAB v3.6 · Blindspot database

10,657

Coding exons lying wholly inside GIAB segmental-duplication or low-mappability intervals on GRCh38, across 2,263 genes. In that sequence a short read may align to another copy instead of this one. On every gene card’s exon map they are red where the sequence is low mappability, which short reads cannot place, and magenta where it is segmental duplication alone, where reads usually still place but a variant matching the other copy can be misassigned.

RefSeq Select · GIAB v3.6 · Blindspot database

Why the home page does not quote Lincoln et al. 2021’s “one in seven” beside these: that figure counts variants that proved technically challenging to detect in 450,000 patients, and ours counts variants that sit in difficult sequence. They measure different things, and side by side they read as a contradiction. Lincoln is below, with the other published evidence.

Where the gene page comes from

Answering “can a test see this gene?” normally means visiting each of these separately and putting the answer together yourself. Blindspot gathers them onto one page, and every flag on it names its source. Each place is counted once, however many of its files are used.

NCBI ClinVar — pathogenic and likely-pathogenic variants, on GRCh38 and lifted to T2T-CHM13
NCBI dbVar (nstd102) — pathogenic deletions and duplications, for the copy-number and zygosity rows
ClinGen — gene–disease validity, dosage sensitivity, actionability and expert-panel variant curations
GenCC — gene–disease submissions and inheritance from every contributing curator
gnomAD — gene constraint
HGNC — approved, previous and alias gene symbols
NCBI RefSeq — gene and transcript models on GRCh38, GRCh37 and T2T-CHM13, including patch and retired models
MANE — Select and Plus Clinical transcripts
Genome in a Bottle (NIST) — difficult-region stratifications, HG002 benchmarks, the challenging medically relevant genes set and assembly-error exclusions
Genome Reference Consortium — the issue records behind GRCh38 fix patches
T2T Consortium — the T2T-CHM13 reference genome
UCSC Genome Browser — assembly gaps, alternate loci and Alu repeats
ENCODE SCREEN — candidate cis-regulatory elements near each gene
Aganezov et al. 2022 — genes affected by GRCh38 false duplications and corrected in T2T-CHM13
Mandelker et al. 2016 — exon-level short-read dead zones in medically relevant genes
Wojcik et al. 2023 — which follow-up method answers which kind of limitation

Not counted: PubMed, which Blindspot uses to verify citations, and the platform’s own curated tables of remedies, limitation types and paralogue cases. Versions and retrieval dates for every source are in the methods.

The scale of the problem

Independent clinical studies show that a meaningful share of pathogenic variants are technically challenging to detect, and that many missed diagnoses were within reach of methods already available. Technical blind spots are one cause among several, and the studies below say which.

Clinical scale & diagnostic impact

1 in 7

In clinical testing of 450,000+ patients, 13.8% of P/LP variants were technically challenging to detect by standard NGS — across segmental duplications, low-complexity regions, large indels, and CNVs (Lincoln et al. 2021, Genet Med).

PMID 34007000

1 in 3

In 4,577 molecularly characterized families, 34.3% involved at least one diagnostic challenge. Most were not sequencing problems — phenotype, pedigree, gene and variant interpretation — and applying those lessons to 314 cases with negative clinical exome or genome reports found the likely cause in 54.5% (AlAbdi et al. 2023, Nat Commun). A reminder that a technical blind spot is one reason among several.

PMID 37644014

UDN

A case series from the NIH Undiagnosed Diseases Network: 12 participants who could have been diagnosed with clinically available methods — phenotype-driven testing, the right tissue, an updated platform, or recognizing alternate transcript nomenclature (Cope et al. 2020, Mol Genet Genomic Med).

PMID 32730690

Caveats

Ten years of clinical whole-exome sequencing surfaced systematic mistakes and caveats — including alternate-contig multi-mapping artifacts and annotation drift as new transcripts are added (Corominas et al. 2022, Hum Mutat).

PMID 35191116

Standards & the call for a gene catalog

Consensus

A 2026 consensus whose authors include Genome in a Bottle (NIST), ClinGen, the Genome Reference Consortium and T2T calls for a maintained catalog of medically important genes annotated with the technical complexities of calling them and the assays shown to resolve them, with every disease gene reviewed to decide whether standard benchmarks are enough or another testing method is needed (Ashley et al. 2026, Nature). Blindspot was begun before this paper appeared and was developed independently of it; the two reach the same need from different directions. It is a computational draft of that kind of catalog — not the expert-adjudicated standard the paper proposes, and its remedies are classes of method rather than assays validated gene by gene.

PMID 42387168 · 10.1038/s41586-026-10621-5

Sequencing methods & dead zones

HiFi

PacBio HiFi/CCS long reads achieve 99.8% accuracy at ~13.5 kb read length, with SNV precision and recall of at least 99.91% against the GIAB HG002 benchmark — the basis for treating long-read sequencing as an orthogonal method where short reads cannot resolve a region (Wenger et al. 2019, Nat Biotechnol).

PMID 31406327

An exome-wide homology catalog in four lists of increasing stringency. Its strictest, the NGS Dead Zone list — exons or long stretches with 100% identity elsewhere in the genome — reaches 67 of the 5,694 panel genes; its high-stringency problem list, 195. Homology is typed as same gene, different functional gene, pseudogene or non-CDS (Mandelker et al. 2016, Genet Med). Blindspot uses it as independent validation of the classification, not as a count of its own.

doi:10.1038/gim.2016.58

Dead zones

The same catalog identifies genes where even Sanger sequencing fails to provide orthogonal confirmation — a reminder that a "confirmatory" method can share the blind spot of the primary assay.

Mandelker et al. 2016

Reference assemblies & liftover

T2T

The first complete, gapless human genome sequence (T2T-CHM13) — the reference against which Blindspot computes assembly comparison and liftover analysis (Nurk et al. 2022, Science).

PMID 35357919

46.9%

2,330 of the 4,964 medically relevant genes in Aganezov et al.’s Table S13 carry the paper’s ClinVar-liftover flag — the flag is theirs, Blindspot counted it. The same paper is the source for GRCh38-vs-CHM13 discordance, false-duplication gene pairs, and the medically-relevant gene impact table (Table S13) that Blindspot draws on (Aganezov et al. 2022, Science).

PMID 35357935

1,688

Of 341,390 pathogenic/likely-pathogenic ClinVar variants, 1,688 (0.49%) do not lift cleanly from GRCh38 to T2T-CHM13 in Blindspot's own Picard LiftoverVcf run — 895 because T2T-CHM13 carries a different reference base at the site, 574 because the position has no T2T-CHM13 counterpart, 218 indels that straddle a break in the liftover chain, and 1 other. Concentrated in 667 genes, led by BRCA1 (84).

NCBI ClinVar · internal pipeline

Reference datasets

99.0%

Of the 19,338 genes with GIAB stratification data on GRCh38, 99.0% (19,150) touch at least one difficult region — which is why the useful question is not whether a gene is difficult, but which category, which exon, and whether an orthogonal method resolves it.

GIAB v3.6 · internal pipeline

GIAB

Genome in a Bottle difficult-region stratification (v3.6) — the segmental-duplication, GC-extreme, low-mappability, and tandem-repeat categories that drive Blindspot's exon-level overlays.

NIST Benchmark

MANE

MANE Select and RefSeq Select transcripts provide the exon / CDS / UTR structure overlaid on difficult regions — so a blind spot can be located to the exon, not just the gene.

NCBI / EMBL-EBI MANE

ClinGen

Gene–disease validity classifications and dosage-sensitivity curation — including refuted associations that flag genes that shouldn't be reported on a given panel.

clinicalgenome.org

ClinVar

The ClinVar 2026-05-30 release underpins Blindspot's pathogenic/likely-pathogenic variant set. The GRCh38→T2T-CHM13 liftover analysis was run against the 2026-06-21 ClinVar release.

NCBI ClinVar

Worked example: a citation that names a gene which no longer exists

Blindspot attaches published evidence to genes. Sometimes the paper and the gene do not line up, because the gene was renamed after the paper was written. ACSM2B is the clearest case on the panel, and it shows why we flag this rather than quietly resolving it.

2003 · Hypertension

A paper studies “MACS2”

Iwai et al. describe an acyl-CoA synthetase family on chromosome 16p12 — SAH, MACS1, MACS2, MACS3 — and associate a Leu513Ser polymorphism in MACS2 with triglyceride level and waist-to-hip ratio.

PMID 12654705 · 10.1161/01.HYP.0000064944.60569.87

2006 · Mol Nutr Food Res

A second paper replicates it, still as “MACS2”

Lindner et al. test the same L513S polymorphism in a Caucasian cohort and report associations with metabolic-syndrome risk factors. Both papers name the gene MACS2, because in 2006 that is what it was called.

PMID 16521160 · 10.1002/mnfr.200500241

20 June 2007 · HGNC

The symbol is split in two

HGNC separates ACSM2 into ACSM2A and ACSM2B. Both of today’s genes list ACSM2 as a previous symbol. Every citation written before this date names a gene that no longer resolves to one entry.

Today

The evidence cannot be assigned by symbol alone

ACSM2B is on the clinical panel; ACSM2A is carried at research tier. Both 2003 and 2006 papers predate the split, so neither states which of the two genes carries L513S. We do not guess, and we do not silently drop the citation — we show it with the ambiguity stated.

Why it happened here

The naming problem and the sequencing problem have one cause

ACSM2B is classified Technology-sensitive (class 4), and 100% of its span lies within a segmental duplication, and 69.1% of its coding sequence is low-mappability. ACSM2A and ACSM2B are a duplicated pair. The sequence similarity that makes short reads hard to place uniquely is the same similarity that kept the two genes under one symbol for years. The literature and the assay stumble over the same fact.

What triggers the caveat — and what does not. We considered flagging citations wherever a gene sits in a segmental duplication, on the theory that duplicated genes attract ambiguous names. Tested across 394 genes, the association is real but modest — roughly a doubling — and 82% of genes with an ambiguous historical symbol are not highly duplicated. Keying the caveat on duplication would have missed most of them. So the trigger is the symbol ambiguity itself, read directly from HGNC previous-symbol records, and duplication appears only as an explanatory clause where it applies, as it does here.

149 genes carry this caveat today, 65 of them on the clinical panel. Seventeen have notes written for them; the rest carry a note generated from HGNC records, which says so. On each, check the coordinates or sequence reported in a cited paper rather than relying on the gene symbol it uses.

What this caveat does not cover. The trigger is shared symbol history, which is narrower than paralogue confusion in general. Where a near-identical paralogue was described later under a new name, no shared symbol ever entered the record and no caveat appears. CBS is that case: it has no previous symbols at all, yet it sits inside the 21q22.3 segmental duplication that carries its near-identical paralogue CBSL, with 100% of its coding sequence inside that duplication. Literature predating the description of CBSL cannot have distinguished the two. Genes in that position are surfaced by the callability classification and the duplication figures instead — the layer built to carry that signal — rather than by widening this trigger onto a proxy.

Built on peer-reviewed science, not vendor claims.

The figures on the home page

Where the gene page comes from

The scale of the problem

Worked example: a citation that names a gene which no longer exists

Trace any flag to its source.