GTEx

GTEx collected 54 non-diseased tissue sites across nearly 1,000 post-mortem human donors. This design enables direct comparison of genetic regulatory effects across tissues within the same individual.

• Sample Size: 948 donors with paired genotype and RNA-seq data
• Tissue Diversity: 31 solid organ tissues, 10 brain sub-regions, whole blood, and cell lines
• Key Insight: A single genetic variant can increase expression in one tissue while having no effect or the opposite effect in another

Each donor was genotyped using whole genome sequencing at 30x coverage, capturing over 80 million genetic variants including single nucleotide polymorphisms, insertions, and deletions.

• Imputation Panel: Variants were imputed using the 1000 Genomes Project reference to expand coverage
• Ancestry: Predominantly European ancestry with representation from African American, Asian, and Hispanic populations
• Quality Control: Variants filtered by minor allele frequency >1% and Hardy-Weinberg equilibrium

RNA sequencing was performed on all collected tissues using a poly-A selection protocol followed by paired-end sequencing at a median depth of 80 million reads per sample.

• Quantification: Gene-level and transcript-level expression measured in TPM and read counts
• Splicing: Exon-exon junction reads enable quantification of alternative splicing events across tissues
• Batch Design: Samples randomized across sequencing batches to minimize technical confounding

The primary analytical output is the catalog of cis-eQTLs—genetic variants within 1 megabase of a gene's transcription start site that significantly associate with its expression level.

• Statistical Model: Linear regression of normalized expression against variant dosage, with PEER factors to correct for hidden confounders
• Discovery: Identified cis-eQTLs for >88% of protein-coding genes in at least one tissue
• Tissue-Sharing: Most eQTLs are shared across multiple tissues, but a subset are highly tissue-specific

GTEx revealed that 48% of eQTLs are shared across all tissues, while the remainder exhibit tissue-specific effects driven by chromatin state differences.

• Enrichment: Tissue-specific eQTLs are enriched in enhancer and promoter regions active in the relevant tissue
• Transcription Factor Mediation: Tissue-specific effects often explained by differential expression of transcription factors that bind the variant
• Disease Relevance: GWAS variants colocalize more frequently with tissue-specific eQTLs in disease-relevant tissues

Genomic loci where genetic variants are statistically associated with the variation in mRNA expression levels of a specific gene. GTEx is the definitive resource for cataloging cis-eQTLs and trans-eQTLs across 49 human tissues, linking regulatory DNA to transcript abundance. This mapping is fundamental for interpreting the downstream functional impact of non-coding variants identified in genome-wide association studies.

A normalization method for RNA-seq data that corrects for both gene length and sequencing depth, allowing direct comparison of transcript proportions across samples. GTEx provides gene expression data in TPM to facilitate cross-tissue analysis. The metric scales the total transcript count to one million, ensuring that a TPM value represents the relative molar concentration of a transcript in the sample.

A computational perturbation method where every nucleotide in an input DNA sequence is systematically mutated to measure the predicted change in a model's output. Applied to models trained on GTEx data, this technique reveals regulatory motif logic by identifying which base pairs are critical for predicted expression. It transforms a black-box predictor into a mechanistic hypothesis generator for variant interpretation.

Systematic non-biological variations introduced by differences in sample processing, reagent lots, or sequencing platforms. GTEx's complex logistics across dozens of tissue collection sites make it susceptible to batch effects that can confound machine learning models. Methods like ComBat-Seq use negative binomial regression to adjust for known technical covariates while preserving the true biological signal of tissue-specific expression.