Skip to content

Reference genome

Species: Nothobranchius furzeri (African turquoise killifish)
Strain: GRZ
Assembly: GCF_043380555.1 NfurGRZ-RIMD1
Source: NCBI RefSeq, December 2024
Genome size: ~1.3 Gb
GTF file: data/GCF_043380555.1_NfurGRZ-RIMD1_genomic.gtf

The GTF is also used by raw_RNAseq_process/run_rnaseq.sh to build the STAR index and quantify gene counts. See Raw RNA-seq Pipeline.

Gene ID Mapping

The KillifishAtlas uses NCBI-style gene names (e.g. actb, mb21d1, LOC107374091). Query datasets produced from the NfurGRZ-RIMD1 genome assembly (NCBI RefSeq GCF_043380555.1) use Ensembl IDs (ENSNFUG00000001234). Bridging these two namespaces requires a multi-source mapping pipeline.

Building the unified map (data/build_gene_map.py)

Script: data/build_gene_map.py

Merges three sources into data/gene_id_map.csv:

Source 1: GCF_043380555.1_NfurGRZ-RIMD1_genomic.gtf
          → gtf_gene_id, gtf_gene_name, ncbi_gene_id

Source 2: ncbi_gene2ensembl_nfurzeri.csv
          → ncbi_gene_id ↔ ensembl_gene_id

Source 3: query_to_atlas_gene_mapping.csv
          → ensembl_gene_id → atlas_gene

Merge strategy:

  1. Parse gene records from the GTF, extracting gene_id, gene name, and GeneID db_xref.
  2. Join with gene2ensembl on ncbi_gene_id to add ensembl_gene_id.
  3. Join with the Atlas mapping on ensembl_gene_id to add atlas_gene.
  4. Supplement with Ensembl IDs in gene2ensembl not reachable via the GTF.
  5. Supplement with Atlas entries not reachable via either route.
cd data/
python build_gene_map.py
# → gene_id_map.csv

Output columns

Column Description
gtf_gene_id RefSeq gene ID from GTF (e.g. gene1234)
gtf_gene_name Gene symbol from GTF (e.g. actb)
ncbi_gene_id NCBI GeneID integer
ensembl_gene_id Ensembl ID (ENSNFUG...)
atlas_gene Atlas gene name used as row index in count matrices

Three-layer mapping in GeneMapper

src/gene_mapping.py applies three layers in order:

Layer Method
1 Direct lowercase gene_name → Atlas
2 BioMart external_gene_name fallback
3 ENSNFUG → GeneID 107XXXXXX → LOC107XXXXXX

Coverage against the Atlas gene symbols -> ENSNFUGxxx

Measured by unittests/test_gene_mapping.py against GSE308970_Counts_Atlas_allbatches_merged_v3.csv

If a query experiment measured all Atlas gene symbols by their ENSNFUG IDs, how many could GeneMapper successfully translate? (25,122 Atlas genes × 677 samples; run 2026-05-26) :

Gene type Atlas total Covered by GeneMapper Coverage
Named symbols (e.g. actb) 10,533 7,706 73.2 %
LOC genes (e.g. LOC107374091) 14,589 4,311 29.5 %
All genes 25,122 12,017 47.8 %

The lower LOC-gene coverage reflects a biological reality: most LOC107XXXXXX entries in the Atlas are unannotated loci that lack Ensembl cross-references in either BioMart or the NCBI gene2ensembl table. The 13,105 uncovered genes are dropped before clock training and have negligible impact on clock accuracy because unannotated loci carry little age-predictive signal.

Coverage: raw_RNAseq_process output (PRJNA817434)

Data produced by raw_RNAseq_process/run_rnaseq.sh uses NCBI-style gene names directly from the GTF — no ENSNFUG IDs, so GeneMapper does not apply. Coverage is measured as a direct set intersection with Atlas gene names.

Measured against raw_RNAseq_process/results/PRJNA817434/PRJNA817434_raw_count.csv (36,530 genes × 9 samples; run 2026-05-26):

Gene type PRJNA817434 total In Atlas Coverage
Named symbols (e.g. acsl4a) 17,657 7,098 40.2 %
LOC107 genes (e.g. LOC107374091) 6,203 5,103 82.3 %
LOC139 genes (e.g. LOC139071432) 7,031 0 0.0 %
tRNA / KEG92 entries 5,639 — excluded
Usable total 36,530 12,249 33.5 %

LOC139XXXXXX genes come from a newer NCBI annotation (GeneID range 139M) that does not exist in the Atlas (which uses the older 107M range). They are unmappable without rebuilding the Atlas with the updated GTF.