Example data¶
The repository ships a compact copy of a real xQTL dataset in examples/mage/data/
(32 MB). Every example in this documentation uses it, so you can run each command as
written right after installation.
What it is¶
- Expression: RNA-seq of lymphoblastoid cell lines (LCLs) from the MAGE study (Multi-ancestry Analysis of Gene Expression), 729 individuals of the 1000 Genomes Project, normalised (TMM, then inverse normal transform per gene).
- Genotypes: the New York Genome Center (NYGC) high-coverage whole-genome sequencing of the 1000 Genomes samples.
- Reference: cis-eQTLs from an independent study, GTEx v10: for each gene with a significant eQTL (an eGene) in EBV-transformed lymphocytes, its sentinel (strongest) variant. GTEx donors are not MAGE individuals.
- Region: chromosomes 21 and 22 only.
| What | Size |
|---|---|
| Individuals (genotypes and expression) | 729 |
| Genotyped variants (MAF ≥ 5%) | 256,645 (127,563 on chr21, 129,082 on chr22) |
| Genes (expression and loci) | 367 |
| Covariates | 66: sex, PC1-5, PEER1-60 |
| Known cis-eQTL pairs | 367 (one per gene; 346 distinct variants) |
Why the reference comes from another study
The known eQTLs were discovered in GTEx donors, not in these 729 individuals. That makes the comparisons meaningful: a reference discovered in the same samples would fit their own noise, make identity matches look stronger than they are and favour the covariate setup used for its discovery. Use an independent reference with your own data too.
Files¶
examples/mage/data/
├── geno/
│ ├── mage.pgen_list.tsv pgen list: which file holds which chromosome
│ ├── mage_chr21.pgen genotypes, chr21 (14.6 MB)
│ ├── mage_chr21.psam sample IDs
│ ├── mage_chr21.pvar.idx.gz indexed variant list (1.3 MB)
│ ├── mage_chr21.pvar.idx.gz.tbi tabix index of the variant list
│ └── mage_chr22.* the same for chr22 (15.2 MB pgen)
├── pheno/
│ ├── mage_expr.chr21_22.regenie.tsv.gz expression, 729 × 367 (0.6 MB)
│ ├── mage_covariates.regenie.tsv.gz covariates, 729 × 66 (0.2 MB)
│ └── mage_genes.chr21_22.bed gene TSS loci (17 KB)
└── reference/
└── gtex_lcl_eqtl.chr21_22.tsv.gz known cis-eQTL pairs from GTEx (12 KB)
Genotypes¶
The pgen list tells qpgen which PLINK 2 file set holds which chromosome. Columns are chromosome, first position, last position and the file prefix (relative prefixes are looked up next to the list):
$ cat examples/mage/data/geno/mage.pgen_list.tsv
21 1 1000000000 mage_chr21
22 1 1000000000 mage_chr22
The psam lists the samples, one ID per line:
The indexed pvar (.pvar.idx.gz with its .tbi) lists the variants with an
INDEX column giving each variant's row in the .pgen. qpgen looks variants up through
the tabix index instead of reading the whole list. anqchor prepare index-pvar makes
one from an ordinary .pvar.
$ gzip -dc examples/mage/data/geno/mage_chr21.pvar.idx.gz | head -5
#CHROM POS ID REF ALT QUAL FILTER INFO INDEX
21 5031747 21:5031747:G:GA G GA . . . 1
21 5031747 21:5031748:GA:G GA G . . . 2
21 5083315 21:5083315:A:C A C . . . 3
21 5083679 21:5083679:G:A G A . . . 4
The .pgen files are binary. The genotypes are unphased hard calls.
Expression¶
A samples × genes matrix in regenie layout: FID, IID, then one column per gene. Gene
IDs are ENSEMBL_ID:GENE_NAME.
$ gzip -dc examples/mage/data/pheno/mage_expr.chr21_22.regenie.tsv.gz | head -3 | cut -f1-5
FID IID ENSG00000275993.3:SIK1B ENSG00000278931.1:bP-2189O9.2 ENSG00000279720.1:SOWAHCP2
HG00096 HG00096 -0.238523 0.224454 2.54496
HG00100 HG00100 -0.217436 -0.309645 -0.0685407
The sample IDs are the genotype IDs, so no sample map is needed.
Covariates¶
The same layout, one column per covariate. sex is text (XX/XY), so it is a
categorical covariate: anqchor keeps the text levels and passes --cat-cov-names sex to
qpgen, which replaces it by one indicator column (XX is the reference level).
$ gzip -dc examples/mage/data/pheno/mage_covariates.regenie.tsv.gz | head -3 | cut -f1-9
FID IID sex PC1 PC2 PC3 PC4 PC5 PEER1
HG00096 HG00096 XY -0.0216243 0.0487524 0.00362148 0.0446186 0.00187148 -0.0565679
HG00100 HG00100 XX -0.0215191 0.0503942 0.00281614 0.040869 0.000118772 0.0722641
The columns after PEER1 are PEER2 to PEER60. With --cov-include you can use a
subset, for example --cov-include sex,PC1-5,PEER1-10 (ranges expand).
Non-numeric columns are categorical automatically. A covariate whose categories are coded as
numbers, for example a sequencing batch 1, 2, 3, would be read as one numeric
covariate; name it with --cat-cov-names batch (config cov_categorical: [batch]) to treat
it as categorical. See covariates.
Gene loci¶
A BED-like file with each gene's transcription start site. Gene IDs are the same as in
the expression file. trans evaluation uses it to tell cis cells from trans cells; cis
evaluation needs it only with --assoc-engine rect.
$ head -4 examples/mage/data/pheno/mage_genes.chr21_22.bed
#chrom start end id
21 6111130 6111131 ENSG00000275993.3:SIK1B
21 8857259 8857260 ENSG00000278931.1:bP-2189O9.2
21 9083100 9083101 ENSG00000279720.1:SOWAHCP2
Reference pairs¶
The known cis-eQTLs: trait, variant (CHROM:POS:REF:ALT), effect size, its standard
error and p-value. Each row is one GTEx eGene with its sentinel variant; beta, se and
pval are GTEx's slope, slope_se and pval_nominal, and beta is the effect of one
copy of the ALT allele. The trait is the gene's ID in the expression file (matched on the
Ensembl ID without its version).
$ gzip -dc examples/mage/data/reference/gtex_lcl_eqtl.chr21_22.tsv.gz | head -4
trait variant beta se pval
ENSG00000275993.3:SIK1B 21:6122794:C:T -0.448959 0.0888711 8.054e-07
ENSG00000278931.1:bP-2189O9.2 21:9370164:G:A 0.484109 0.10423 5.31901e-06
ENSG00000279720.1:SOWAHCP2 21:9088050:G:A 0.708996 0.180329 0.000107185
Every reference variant is present in the genotypes, so all 367 pairs are tested. The GTEx p-values range from 1e-102 to 1e-4 (median 8e-11): some eGenes are strong, others barely significant in GTEx, as in any real reference.
What the example shows¶
- Identity matching: 727 of 729 samples are confirmed (14 only leniently). The
expression sample labelled
NA11919does not match its own genotypes (rank 253 of 729), and no other individual matches it clearly either, so the result is "unclear" rather than a named mismatch. BothNA11919and its best (but weak) candidate,HG00237, are in the data (genotypes and expression), so you can inspect the case yourself (identity matching). - cis evaluation: the known pairs replicate much better with PEER factors: 178 of 367 at p < 10⁻⁵ without covariates, 251 with all 66 covariates (cis evaluation).
How it was made¶
examples/mage/make_compact_data.sh rebuilds examples/mage/data/ from the full MAGE
and NYGC files and the GTEx v10 eGene table for EBV-transformed lymphocytes
(Cells_EBV-transformed_lymphocytes.v10.eGenes.txt.gz, part of the single-tissue cis-eQTL
data on the GTEx portal). It needs plink2, anqchor and Python with pandas:
FULL_DATA_DIR holds the MAGE and NYGC files, GTEX_EGENES is the path of the GTEx eGene
table and PLINK2 the plink2 binary (default: plink2 on your PATH). For example:
bash examples/mage/make_compact_data.sh /data/mage \
/data/gtex/Cells_EBV-transformed_lymphocytes.v10.eGenes.txt.gz
What it keeps, and why:
| Part | Kept | Why |
|---|---|---|
| Genotypes | the 729 individuals with expression; every variant with MAF ≥ 5% among them; unphased; psam with #IID only; pvar reduced to CHROM POS ID REF ALT, then indexed |
any common variant can be queried, not just the reference variants, so trans rectangles and your own variant lists work |
| Reference | the GTEx eGenes on chr21/22 with q ≤ 0.05 whose sentinel variant is in those genotypes (same CHROM:POS:REF:ALT) and whose gene is in the MAGE expression data (same Ensembl ID, ignoring the version); columns trait variant beta se pval |
an independent reference in the same cell type, and every pair can be tested |
| Expression | the 729 samples × the genes of those reference pairs; values rounded to 6 significant digits | the analyses only use genes with a known pair |
| Gene loci | the TSS of those genes, with the expression IDs and the genotypes' chromosome names (21, not chr21) |
IDs and chromosomes match the other files without any mapping |
| Covariates | sex, PC1-5, PEER1-60 for the 729 samples, regenie layout, 6 significant digits | the covariates used by the MAGE study |
Unused columns (the other columns of the GTEx table; the extra pvar columns) are dropped.
Provenance of the full files
The script starts from the NYGC phased chr21/22 genotypes of 3,202 1000 Genomes
samples, the MAGE expression matrix of 729 samples × 20,154 genes with its tensorQTL
BED, the MAGE covariate table (one row per covariate) and the GTEx v10 eGene table of
EBV-transformed lymphocytes. GTEx tests 913 genes on chr21/22 and calls 516 of them
eGenes at q ≤ 0.05. Of those 516, 367 are kept: 72 genes are not in the MAGE expression
data, 67 have a sentinel variant with MAF below 5% in the 729 individuals, and 10 have
a sentinel variant that the genotype release does not contain under the same
CHROM:POS:REF:ALT.