allelix-transcripts / README.md
dial481's picture
Transcript caches for Allelix 2.8 (MANE v1.5, GENCODE 50)
61c8d82 verified
|
Raw History Blame Contribute Delete
4.81 kB
---
license: cc0-1.0
task_categories:
- tabular-classification
tags:
- genomics
- transcripts
- mane
- gencode
- variant-effect
- GRCh37
- GRCh38
- bioinformatics
- sqlite
pretty_name: Allelix Transcript Caches
size_categories:
- 10K<n<100K
---
# Allelix transcript caches
Pre-built transcript caches for [Allelix](https://github.com/dial481/allelix),
which reads them to give each variant's gene effects on MANE transcripts.
The transcript and reference-sequence caches Allelix 2.8 reads to give each
variant's gene effects (its `CSQ` field and the report's `consequence`). Allelix
downloads them with `allelix db update`, pinned to this dataset's commit and to
each file's SHA-256, and checks both before use.
## Files
| File | What it holds |
|---|---|
| `transcripts.GRCh38.sqlite.gz` | GRCh38: the MANE v1.5 transcripts (MANE Select and MANE Plus Clinical) on the primary assembly, and Ensembl 116's 37 mitochondrial transcripts, with their exons, CDS and translation edits; every GENCODE 50 basic gene's span; the build's checks and its input files. |
| `transcript_windows.GRCh38.sqlite.gz` | GRCh38: each transcript's exons with 50 bases either side, cut from the reference; it names the transcript cache it was cut for by SHA-256. |
| `transcripts.GRCh37.sqlite.gz` | GRCh37: the MANE v1.5 transcripts that GENCODE 50's lift37 annotation carries over whole and that match MANE's sequences on hg19, and the same 37 mitochondrial transcripts. |
| `transcript_windows.GRCh37.sqlite.gz` | GRCh37: the same windows, from hg19 (MT from the rCRS). |
| `transcripts.GRCh37.dropped.tsv` | GRCh37: each MANE transcript left out, and why. Not read by Allelix. |
## Sources and pinned versions
- **MANE v1.5** (NCBI and EMBL-EBI): `MANE.GRCh38.v1.5.ensembl_genomic.gtf.gz`,
`MANE.GRCh38.v1.5.summary.txt.gz`, `MANE.GRCh38.v1.5.ensembl_rna.fna.gz` and
`MANE.GRCh38.v1.5.ensembl_protein.faa.gz`, from
https://ftp.ncbi.nlm.nih.gov/refseq/MANE/MANE_human/release_1.5/. The RNA and
protein sequences are used only to check each transcript.
- **Ensembl release 116**, mitochondrial genes:
`Homo_sapiens.GRCh38.116.chromosome.MT.gff3.gz` from
https://ftp.ensembl.org/pub/release-116/gff3/homo_sapiens/ (SHA-256
`483cb47af5b9ee945b29c6ed2f794cc3625b87a125f73fdf0fdd1b9e47d663f3`).
- **GENCODE release 50** (Ensembl 116, the release MANE v1.5 names):
`gencode.v50.basic.annotation.gtf.gz` for GRCh38's gene spans, and
`gencode.v50lift37.basic.annotation.gtf.gz` from the release's
`GRCh37_mapping` directory for GRCh37, both from
https://ftp.ebi.ac.uk/pub/databases/gencode/Gencode_human/release_50/ and
checked against its `MD5SUMS`.
- **Reference sequence**: UCSC's `hg38.2bit` and `hg19.2bit` (GRC assemblies),
from https://hgdownload.soe.ucsc.edu/goldenPath/, checked against their
`md5sum.txt`. MT on both builds is the rCRS, hg38's `chrM`: hg19's own chrM is
not.
Each cache's `source_files` table records every input's name, SHA-256 and
size, and its `build_checks` table the transcripts kept and dropped by reason.
## How a transcript is kept
- **GRCh38:** every MANE v1.5 transcript on the primary assembly (64 MANE
Select transcripts on alt or patch scaffolds only are left out), whose
spliced sequence equals MANE's RNA and whose translation equals MANE's
protein, with the translation's own edits read from that protein: an
initial Met where codon 1 is not ATG, selenocysteine at a TGA, and a CDS
with no stop (TMEM247).
- **GRCh37:** a lift37 transcript stands for its MANE transcript only with
MANE's ENST and version, its MANE tag, a whole-contig remap
(`remap_status full_contig`), the same exons and CDS length, and the same
sequence checks on hg19. Every other is dropped and listed in
`transcripts.GRCh37.dropped.tsv`; Allelix gives its variants no entry
rather than a wrong one.
## Build
From the Allelix repository at the commit that pins these files:
```
python scripts/build_transcript_cache.py --build GRCh38 --mane-dir MANE --reference hg38.2bit \
--mt-gff3 Homo_sapiens.GRCh38.116.chromosome.MT.gff3.gz \
--gencode-gtf gencode.v50.basic.annotation.gtf.gz \
--output-dir OUT --source-date 2026-11-20 --allow-drops 0 --gzip
python scripts/build_transcript_cache.py --build GRCh37 --mane-dir MANE --reference hg19.2bit \
--mt-reference hg38.2bit --mt-gff3 Homo_sapiens.GRCh38.116.chromosome.MT.gff3.gz \
--gencode-gtf gencode.v50lift37.basic.annotation.gtf.gz \
--output-dir OUT --source-date 2026-11-20 --allow-drops 2000 --gzip
```
The same inputs and `--source-date` give the same bytes and SHA-256s.
## Licences
- MANE: NCBI and EMBL-EBI, public, with no restriction on reuse.
- GENCODE and Ensembl: EMBL-EBI, free use.
- The reference sequence: the GRC assemblies via UCSC, freely redistributable.