CRISPR Genome Off Target Map

Where else could this guide cut? The whole genome, counted

Every guide RNA the US public substance registry publishes — found by its scaffold, not by a name anyone remembered — screened against every NGG site on both strands of the human genome, exactly. 304,796,751 candidate sites, one integer each. No sampling, no seed heuristic, no alignment score, and no parameter to choose.

The question, and why it outlives the trial

Casgevy — exagamglogene autotemcel — is an approved CRISPR therapy for sickle cell disease and β-thalassemia. Others here are late-stage in-vivo therapies and CAR-T manufacturing guides. The trials decide whether they work. None of them answers the question a patient would ask about a molecule that edits their genome, and that question does not expire:

Besides its intended cut site, where else in my genome could this guide direct a cut?

That is a discrete question, so it is counted rather than estimated.

The rule

SpCas9 cuts where two conditions hold together: a PAM — three bases immediately 3′ of the protospacer matching NGG — and sufficient complementarity between the 20-base spacer and the protospacer. Bases are integers; a position matches when the two codes are equal.

The PAM is what makes this tractable. A 20-mer against three billion bases would have three billion positions to consider on each strand; requiring NGG discards fifteen sixteenths of that space before any comparison happens, leaving 304,796,751 real candidate sites. Every one is examined, on both strands:

forward at i : protospacer = g[i ..< i+20],  PAM = g[i+20 ..< i+23], g[i+21]=G, g[i+22]=G
reverse at i : PAM read on the other strand — g[i]=C, g[i+1]=C
               protospacer = reverseComplement(g[i+3 ..< i+23])

N bases occur in long runs in the assembly. A window containing any N is not scored and is counted separately — absence is not a match, and it is not a mismatch either. 1,718 windows were set aside this way.

Why the scope is the registry and not a list

The first version of this screen carried two guides, found by searching the registry for drug names recalled from memory. That is a completeness defect no harness can catch: a guide nobody recalls is silently absent from a safety map, the arithmetic correct and the coverage wrong.

So all 742 substances of class nucleicAcid were scanned for the canonical SpCas9 sgRNA scaffold GUUUUAGAGCUAGAAAUAGCAAGU. Fifteen carry it, and in every one the scaffold begins at position 20 — which is what makes the 20-base spacer a measurement rather than a convention.

Finding guides by structure rather than by name vindicated an earlier refusal instead of overturning it. NTLA-2001's own record is the full 4,423-base molecule carrying the Cas9 messenger RNA, with no isolatable spacer, and it was refused. Its guide exists under a separate record — Nexiguran, UNII 5G537B4BTJ — and the structural scan found it. Refusing to guess at the mRNA record lost nothing.

The cut site is measured, not declared

Most guide records carry no target annotation, so a screen that demands a declared target chromosome cannot enumerate a registry — it would have reported nothing for fourteen of the fifteen. A guide's measured cut site is therefore where its zero-mismatch match actually falls in the assembly. Off-targets are the sites outside it. A guide with no zero-mismatch site anywhere gets no off-target list, because its published spacer and the assembly disagree.

The rule was validated on chromosome 2 before the whole-genome run: only Casgevy, whose BCL11A target is on chr2, found a site there, and all fourteen others correctly found none.

What the screen found

guides screened   : 15
sequences scanned : 194
bases scanned     : 3099750718
NGG PAM sites     : 304796751
windows with N    : 1718

guide	UNII	perfect sites	measured cut site	1mm	2mm	3mm	4mm	sites ≤4mm
EVONCABTAGENE-PAZURGEDLEUCEL-SINGLE-G	EQW8RVL4CV	1	chr15 (1 site)	0	0	15	457	472
EVONCABTAGENE-PAZURGEDLEUCEL-SINGLE-G	FKP72X9XKK	1	chr14 (1 site)	0	0	37	279	316
EXAGAMGLOGENE-AUTOTEMCEL-GUIDE-RNA-SE	L28RZ5CC6K	1	chr2 (1 site)	0	0	6	137	143
Lonvoguran	D8UQ4B2T7M	1	chr4 (1 site)	0	0	4	182	186
Nenzinacogene-autogeleucel-gRNA-targe	YGA7BAF735	1	chr3 (1 site)	0	1	7	121	129
Nexiguran	5G537B4BTJ	1	chr18 (1 site)	0	0	18	178	196
Ristoglogene-autogetemcel-gRNA	5UBM9CGH6K	2	chr11 (2 sites)	0	0	10	79	89
Soficabtagene-geleucel-single-gRNA-ta	ENS57C5JUZ	1	chr17 (1 site)	0	0	2	61	63
Soficabtagene-geleucel-single-gRNA-ta	93A4Y2S6E2	1	chr14 (1 site)	0	0	5	136	141
TACATRESGENE-AUTOLEUCEL-GUIDE-RNA-(GR	3KQV6T97QD	2	chr7 (2 sites)	0	0	2	69	71
TGFBR2-5-sgRNA-(zugocabtagene-geleuce	GPK7BXF67W	1	chr3 (1 site)	0	0	6	83	89
Taziguran	A2N98QL2SL	1	chr4 (1 site)	0	1	18	189	208
Tremtelectogene-empogeditemcel-Guide-	B6ZZE44GUB	1	chr19 (1 site)	0	0	15	105	120
VOLAMCABTAGENE-DURZIGEDLEUCEL-SINGLE-	4M5F9ZC9EH	1	chr19 (1 site)	0	0	0	47	47
ZC3H12A-10-sgRNA-(zugocabtagene-geleu	RC77WK8XEG	1	chr1 (1 site)	0	0	3	32	35

The seal

MARKER  CRISPR_GENOME_OFFTARGET_EXACT__COMPLETE_ENUMERATION_IS_OBSERVER_INVARIANT
sha256  487b4f81de2d24bd0bb11ecd1d8d42778e3a5d91b9edb33627c86dcc8df34980

The assembly is GENCODE GRCh38 primary, pinned by digest so a reader can confirm they hold the same bytes: b760d18dbb651dd14dfc290083371b3ef3bff122d43a9cefb13ca4ecf38f05ca.

Reproduce

git clone https://github.com/gaiaftcl-sudo/uum8dSolarResearch.git
cd uum8dSolarResearch
swiftc -O reproduce/crispr-genome-offtarget-exact.swift -o /tmp/crispr
curl -sL https://ftp.ebi.ac.uk/pub/databases/gencode/Gencode_human/latest_release/GRCh38.primary_assembly.genome.fa.gz \
  | gunzip -c | /tmp/crispr corpus/crispr-atlas/guides_all.tsv

No account, no key, no data-use agreement, and no floating point anywhere in the exact path.

The verdict

Stated plainly, because a safety map that ends in hedging is not a safety map.

What this screen DOES call. Every one of the fifteen guides has exactly one perfect match in the genome and zero sites at a single mismatch; thirteen of fifteen have zero at two. On the question this instrument can answer — how many places in the genome match this guide, exactly — these are clean guides, and the count is not an opinion. Casgevy, an approved therapy people are alive because of today, is among the cleanest.

What this screen does NOT call, and nobody should read into it. Affine.Earth does not call any of these therapies safe. A site at three or four mismatches is a place the chemistry could direct a cut; whether it does, in a cell, at a dose, in that chromatin state, is a laboratory question this program has not asked and cannot answer. Off-target potential is one input to safety among many, and this page measures only that one.

Where a bench should look. The sites listed at ≤4 mismatches, with their coordinates and strands, are where a laboratory would start if it wanted to check a guide experimentally — and they are published in full for exactly that reason. Nothing here is medical advice and nothing here should change anyone's treatment.

What the page provides is the exact, complete, re-derivable enumeration any such conversation should start from, available to anyone without permission and without trusting us.

Related

• The exact off-target atlas of the nucleic-acid medicines — the same discipline over the transcriptome.
• Cures without the gatekeeper
• Study 26 — Master regulator bonds

Rights — source-available, not open-source

This wiki and its programs are published source-available: the source is visible so anyone can inspect it and re-derive every figure. That visibility grants no rights. The repository carries no LICENSE, which under default copyright means all rights are reserved. Any other use requires a separate written licensing agreement with the authors.