The Dataset Viewer has been disabled on this dataset.

CRISPRme+ 2.10 indexes β€” index format 2

The precomputed indexes of CRISPRme+ 2.10.0 and later, built by CRISPRitz 2.10.0 in index format 2, with each variant panel's Dictionaries and genotype store once, and the enriched-genome kits they were built from. crisprme.py download --what index --index-name <name> fetches an index from here, with what it needs, for any name of this repo.

Why this is a separate repo

These indexes are not readable by earlier releases. An index of format 2 opens and ends with a magic and records its build parameters in a header; CRISPRme+ 2.8.1 refuses one ("this CRISPRme reads index format 1 only") and 2.8.0 aborts on one. The repos earlier releases read, lucapinello/crisprme-data and lucapinello/crisprme-data-general-index, keep their archives (index format 1) unchanged, so an older client never finds an index here. CRISPRme+ 2.10.0 reads both formats; a download of an archive of 2.8.1 by its own name names the index here that replaces it.

Versions: CRISPRme+ and CRISPRitz share their major.minor: CRISPRme+ 2.10.z runs on CRISPRitz 2.10.z (Bioconda pins crispritz >=2.10.0,<2.11, and CRISPRitz 2.10.0's recipe asks for crisprme >=2.10.0; the Docker image builds the exact CRISPRitz it was tested with, v2.10.0). A breaking change in either package (an index format, the enricher's output, a command line or a file one reads of the other) bumps the minor of both; any other change bumps only the patch of the package it changes (version policy). So every index here serves CRISPRme+ 2.10.z, and a change of the index format would bump the minor of both packages. Every install route of 2.10.0 installs the matching CRISPRitz.

Names

An index is named for its guide length and the side of its PAM, in the order of the target on the strand: 20-NRR_3_hg38 is a 20-nt guide with a 3β€² PAM, TTTN-23_3_hg38 a 23-nt guide with a 5β€² PAM; +<dataset> marks a variant index (20-NRR_3_hg38+hg38_HGDP1kGP). 3 is N, a budget of up to 2 DNA and 2 RNA bulges, the web's ceiling. The 2.8.1 names (NRR_3_hg38, ...) say neither length nor side.

Indexes

Index Guide PAM Reference HGDP + 1000 Genomes HPRC mega
SpCas9 NRR (NAA, NAG, NGA, NGG; the default) 20 nt 3β€² 20-NRR_3_hg38 20-NRR_3_hg38+hg38_HGDP1kGP 20-NRR_3_hg38+hg38_HPRC 20-NRR_3_hg38+hg38_mega
SpCas9 NGG (the fast canonical search) 20 nt 3β€² 20-NGG_3_hg38 20-NGG_3_hg38+hg38_HGDP1kGP
SpCas9 NRG (NAG + NGG) 20 nt 3β€² 20-NRG_3_hg38 20-NRG_3_hg38+hg38_HGDP1kGP
General (PAM-less, 8-nt flank: any 3β€² PAM of up to 8 nt, pruned) 20 nt (21 at 1 + 1) 3β€² 20-NNNNNNNN_3_hg38 20-NNNNNNNN_3_hg38+hg38_HGDP1kGP
SaCas9 NNGRRN (serves NNGRRT, pruned) 22 nt 3β€² 22-NNGRRN_3_hg38 22-NNGRRN_3_hg38+hg38_HGDP1kGP
Cas12a TTTN (serves TTTV, pruned) 20 nt 5β€² TTTN-20_3_hg38 TTTN-20_3_hg38+hg38_HGDP1kGP
Cas12a TTTN 23 nt 5β€² TTTN-23_3_hg38 TTTN-23_3_hg38+hg38_HGDP1kGP

An index serves a search of a PAM its own holds, pruned (an NRR index serves NGG, NAG, NGA and NRG; NNGRRN serves NNGRRT; TTTN serves TTTV), and the general index any 3β€² PAM of up to 8 nt. NRG on HPRC and mega is served by the NRR index there, pruned: the same rows, more slowly than a dedicated NRG index (about twice the wall time in the engine's search of chr22's reference sequence). HGDP + 1000 Genomes is genotyped and phased (4,091 samples), HPRC genotyped and phased (232 assembly-derived genomes; HPRC Release 2's Minigraph-Cactus graph v2.1, hprc-v2.1-mc-grch38, where 2.8.1's HPRC archives are of v2.0), mega sites-only (5 sources, allele frequencies only).

Contigs. The reference indexes and those of HGDP + 1000 Genomes hold all 455 sequences of hg38 (its alt, random and unplaced contigs and chrM included), so a variant search finds every site a reference search finds, but the Cas12a TTTN indexes (TTTN-20_3_*, TTTN-23_3_*), which hold 454: chrUn_KI270539v1 (993 bp, GC 0.80) has no TTT and no AAA, so no TTTN site and no tree, and a TTTN search has nothing to find there. Those of HPRC and mega hold its 24 chromosomes, chr1-22, X and Y. An alt contig repeats a region of a primary chromosome (the HLA region, for one) and no VCF of these panels covers it, so a site there can be reported twice, the second time as a reference row on the alt contig.

What these indexes hold. Every allele of a split multi-allelic site is coded (the four-allele symbol Q where the alleles cover A, C, G and T), each indel is searched in a fake-indel window of 30 reference bases on either side (25 and 27 in 2.8.1's archives), and an index and its Dictionaries hold the records whose FILTER is PASS or ., those CRISPRitz's enricher reads. The mega is rebuilt from the same merged VCFs: its registry and indel frequencies hold exactly the records its genome codes, and each indel carries its AF_max.

Searching them. The default budget is 4 mismatches + 1 DNA + 1 RNA bulge. Above 1 + 1 a search of a variant index here warns: the variant search of the densest windows (the immunoglobulin loci) grows by orders of magnitude, to hours or out of memory; raise the mismatches first (CRISPRme 2.10.1 is planned to search the samples' own haplotypes there, with these indexes as they are). A search of mega withholds by default its windows of 2^20 sequences or more, listing them in the dense-region files (--no-sites-only-density-guard analyses them), and ranks its combinations whose frequencies fail the safeguards' bound after the others.

NNN_3 is not built again for 2.10.0. The general index serves every search NNN_3 served, at the same speed (a 3-nt and an 8-nt PAM-less index cost the same: every position is a site), with the same rows; 2.8.1's NNN_3 archives stay in lucapinello/crisprme-data.

The list, with how each archive is built and published, is the index set, seq_script/rebuild/index_set_2.10.0.tsv.

Archives and sizes

Each archive is a .tar.gz (the table gives its path without it). GB = 10⁹ bytes: each index and companion as publish-index packs it, measured before the publish, and each kit as built; indexes/SHA256SUMS gives each index's and companion's SHA-256, from the Hub, and build-inputs/<dataset>.SHA256SUMS each kit's.

Archive GB Archive GB
indexes/20-NRR_3_hg38 20.17 indexes/20-NRR_3_hg38+hg38_HGDP1kGP 27.84
indexes/20-NGG_3_hg38 3.81 indexes/20-NGG_3_hg38+hg38_HGDP1kGP 5.34
indexes/20-NRG_3_hg38 8.89 indexes/20-NRG_3_hg38+hg38_HGDP1kGP 12.24
indexes/20-NNNNNNNN_3_hg38 82.21 indexes/20-NNNNNNNN_3_hg38+hg38_HGDP1kGP 110.91
indexes/22-NNGRRN_3_hg38 5.41 indexes/22-NNGRRN_3_hg38+hg38_HGDP1kGP 7.52
indexes/TTTN-20_3_hg38 2.83 indexes/TTTN-20_3_hg38+hg38_HGDP1kGP 4.54
indexes/TTTN-23_3_hg38 3.09 indexes/TTTN-23_3_hg38+hg38_HGDP1kGP 4.83
indexes/20-NRR_3_hg38+hg38_HPRC 29.55 indexes/20-NRR_3_hg38+hg38_mega 24.11
indexes/dictionaries-bundle_hg38_HGDP1kGP 16.42 indexes/genotypes_hg38_HGDP1kGP 5.87
indexes/dictionaries-bundle_hg38_HPRC 2.44 indexes/genotypes_hg38_HPRC 0.56
indexes/dictionaries-bundle_hg38_mega 1.95 indexes/indel_genotypes_hg38_HGDP1kGP 7.64
indexes/indel_genotypes_hg38_HPRC 0.59 build-inputs/hg38_HGDP1kGP 1.24
build-inputs/hg38_HPRC 1.18 build-inputs/hg38_mega 1.02

A panel's first index is fetched with its Dictionaries bundle and genotype store, once: the default 20-NRR_3_hg38+hg38_HGDP1kGP with them is 50.13 GB, HPRC's 32.55 GB, the mega's 26.06 GB (its bundle alone). A variant index also brings its PAM's reference index where that is not installed; download --what all installs the default's, 20-NRR_3_hg38.

Provenance

Every archive here was built on 2026-11-14 and 2026-11-15 by CRISPRitz 2.10.0, from one image and one commit, and checked before the publish: the integrity of every index, the panels' counts recounted from their VCFs, and the kits' checksums and member lists (Gate A), the real-data indel checker on genome-wide searches of each panel, the --prune-pam equivalence and the timing against 2.8.1. Their record β€” the build image's sha256 and binaries, each input with its version and checksum (hg38's 455 FASTAs; HGDP + 1000 Genomes' re-phased callset; HPRC v2.1's VCF, its S3 ETag and the split files' SHA-256; the mega's merged VCFs; the samplesID lists), each archive's threads, wall time, memory and size, and each kit's SHA-256 and member list β€” is docs/INDEX_PROVENANCE.md (2.10.0), and the methods of the rebuild are METHODS Β§11. The known limits of 2.10.0, which these indexes share, are in METHODS Β§10.

Layout

crisprme_plus_indices/
β”œβ”€β”€ indexes/
β”‚   β”œβ”€β”€ 20-NRR_3_hg38.tar.gz                           # a reference index: the index alone
β”‚   β”œβ”€β”€ 20-NRR_3_hg38+hg38_HGDP1kGP.tar.gz             # a variant index: it and its _INDELS index
β”‚   β”œβ”€β”€ ...                                            # (the indexes above)
β”‚   β”œβ”€β”€ dictionaries-bundle_hg38_HGDP1kGP.tar.gz       # the panel's Dictionaries, once
β”‚   β”œβ”€β”€ dictionaries-bundle_hg38_HGDP1kGP.json         # what the bundle holds (its fingerprint)
β”‚   β”œβ”€β”€ genotypes_hg38_HGDP1kGP.tar.gz (+ .json)       # the panel's Tier-1 genotype store, once
β”‚   β”œβ”€β”€ indel_genotypes_hg38_HGDP1kGP.tar.gz (+ .json) # its phased indel genotypes, for the genotype oracle
β”‚   β”œβ”€β”€ ... (the same for hg38_HPRC; hg38_mega has a bundle, no genotype store)
β”‚   └── SHA256SUMS                                     # one SHA-256 per .tar.gz
β”œβ”€β”€ build-inputs/
β”‚   β”œβ”€β”€ hg38_HGDP1kGP.tar.gz (+ .SHA256SUMS)           # the enriched-genome kits these indexes were built from
β”‚   β”œβ”€β”€ hg38_HGDP1kGP.members.md5                      # the md5 of each file the kit holds, by its path in it
β”‚   β”œβ”€β”€ hg38_HPRC.tar.gz (+ .SHA256SUMS, .members.md5)
β”‚   β”œβ”€β”€ hg38_mega.tar.gz (+ .SHA256SUMS, .members.md5)
β”‚   └── hg38_HPRC/                                     # HPRC's normalised VCFs, which its index was built from
β”‚       β”œβ”€β”€ hprc.chr1.norm.vcf.gz (+ .tbi)             # one a chromosome: chr1-22, X and Y
β”‚       β”œβ”€β”€ ...
β”‚       └── SHA256SUMS                                 # one SHA-256 per VCF and .tbi
└── README.md

The kits and their inputs. A kit (build-inputs/<dataset>.tar.gz) holds the enriched genome and the fake-indel genome of its panel, Genomes/hg38+<dataset>/ and Genomes/hg38+<dataset>_INDELS/, from which build-index-only builds an index of any PAM of the panel without its VCFs (crisprme.py download --what build-inputs --dataset <dataset>). Its member list, <dataset>.members.md5, gives the md5 of each file the kit holds: the tarball is not byte-reproducible (tar records each file's time, owner and listing order), so its SHA256SUMS says that a download is the published file, and the list what the kit holds, the same for any packing of the same files. build-inputs/hg38_HPRC/ holds the VCFs the HPRC index reads: HPRC Release 2's Minigraph-Cactus graph v2.1 on GRCh38 (hprc-v2.1-mc-grch38, the VCF decomposed with vcfwave), split per chromosome and normalised as the panel's recipe does it, hprc.<chr>.norm.vcf.gz and its .tbi (lucapinello/crisprme-data's vcfs/hg38_HPRC holds the v2.0 files 2.8.1's HPRC archives were built from). HGDP + 1000 Genomes' VCFs are lucapinello/crisprme-data's vcfs/hg38_HGDP1kGP, as they are, and mega's index is rebuilt from that repo's vcfs/hg38_mega.

Each panel's Dictionaries once. A variant archive holds the index, its _INDELS index and a manifest.json, which names the panel's Dictionaries bundle (dictionaries-bundle_<dataset>: the Tier-0 registry, the indel logs, the phased indel genotypes, mega's indel allele frequencies, and the samplesID lists) and its genotype store (genotypes_<dataset>) by name and fingerprint. download fetches each once per panel: a second index of the panel finds them installed, the same, and fetches its own archive (and its PAM's reference index where that is not installed: 20-NGG_3_hg38 for 20-NGG_3_hg38+hg38_HGDP1kGP), not the bundle and the store again. --no-genotypes leaves the genotype store out. (The archives of 2.8.1 each carried their own copy of the Dictionaries, 16.5 GB for every HGDP + 1000 Genomes archive.)

Usage

# the genome, annotations, PAMs and sample lists, and the default reference index
# (20-NRR_3_hg38, from here)
crisprme.py download --what all --path .

# the recommended variant-aware default: phased HGDP + 1000 Genomes, SpCas9 NRR
crisprme.py download --what index --index-name 20-NRR_3_hg38+hg38_HGDP1kGP --path .

# another index of the same panel: its archive, and 20-NGG_3_hg38 if not installed (the
# bundle and store are installed already)
crisprme.py download --what index --index-name 20-NGG_3_hg38+hg38_HGDP1kGP --path .

Two installs of one panel must read one build of its Dictionaries: install an index of 2.8.1 (the older names) in another working directory. A download refuses to put an archive of 2.8.1's Dictionaries over a bundle of this repo, and says which installed indexes read Dictionaries of another build where a bundle replaces them, and where the panel's enrichment in place (Genomes/hg38+<panel>, which 2.8.1's build or kit left) is of another build. A search with no bulges scans it in place of the index where it holds every chromosome of the genome (as 2.8.1's build of all of hg38 left it; its kits hold chr1-22, X and Y alone), and a build of an index reads it: each refuses it, so remove it, or put this repo's kit of the panel in its place.

Verifying

  • An index or a companion: from inside a downloaded indexes/, sha256sum -c --ignore-missing SHA256SUMS.
  • A kit: beside the downloaded tarball, sha256sum -c <dataset>.SHA256SUMS; once unpacked, md5sum -c <dataset>.members.md5 in the folder it was unpacked in (the one that holds its Genomes/).
  • HPRC's VCFs: from inside a downloaded build-inputs/hg38_HPRC/, sha256sum -c --ignore-missing SHA256SUMS.
Downloads last month
45