Expression QTL (eQTL) Resources#

Expression quantitative trait loci (eQTL) identify genetic variants that influence gene expression levels measured by bulk or single-cell RNA sequencing. This resource provides eQTL summary statistics, fine-mapping results, TWAS/qTWAS pre-trained models, and colocalization analyses across multiple brain regions, peripheral tissues, and AD-relevant cohorts.

Overview#

eQTL mapping was performed using the FunGen-xQTL pipeline across diverse human brain regions and cell types. All datasets underwent harmonized genotype QC, expression normalization, and covariate adjustment. Fine-mapping was performed using SuSiE-RSS to generate posterior inclusion probabilities (PIPs), credible sets (CS), and effect sizes. TWAS weight models and quantile TWAS (qTWAS) models are provided for key datasets.

Available Datasets#

Brain Tissue — ROSMAP Cohort#

Brain Tissue — Other Cohorts#

Blood / Peripheral Tissue#

Analyses Performed#

Single-Context Fine-Mapping#

Fine-mapping using SuSiE-RSS is performed per dataset and per molecular phenotype, yielding:

• Posterior inclusion probabilities (PIPs)
• 95% credible sets (CS)
• Standardized effect sizes (beta) and standard errors

Multi-Context Fine-Mapping#

For ROSMAP (DLPFC, PCC, AC) and MSBB cohorts, mvSuSiE (multivariate SuSiE with MASH prior) integrates signals across multiple brain regions jointly to improve power and resolution. Multi-context results are deposited in syn69670592.

Multi-Gene Fine-Mapping#

Separate multi-gene fine-mapping analyses for eQTLs jointly model nearby genes to resolve signals with shared genetic variants. Results available at syn69670592.

Trans-eQTL Fine-Mapping#

Genome-wide trans-eQTL analyses identify distant regulatory relationships (>5 Mb or inter-chromosomal). Trans-eQTL fine-mapping results for ROSMAP DLPFC are provided at ROSMAP DLPFC.

TWAS / qTWAS Models#

Pre-trained transcriptome-wide association study (TWAS) models are available for integrating eQTL weights with AD GWAS summary statistics:

• TWAS weight models (syn69670600): ROSMAP DLPFC, PCC, AC; MSBB; Knight ADRC
• qTWAS models (syn69670611): quantile regression models capturing non-linear eQTL effects
• cTWAS inputs (syn70095142): formatted for causal TWAS accounting for LD structure

Colocalization#

Multi-context colocalization using ColocBoost across cohorts and brain regions identifies shared genetic signals between eQTLs and other molecular traits (syn69670597). AD GWAS–eQTL colocalization results are available at syn69865816 and syn69670630.

Single-Cell eQTL Prediction (scEEMs)#

The scEEMs model predicts cell-type-specific eQTL effects using a CatBoost binary classifier trained on 4,839 genomic features (TSS distance, ABC scores, baseline annotations, cell-type-specific annotations, deep learning variant effect predictions, and gene features) with leave-one-chromosome-out cross-validation. Models trained on ROSMAP snRNA-seq pseudo-bulk data are described in ROSMAP snRNA-seq pseudo-bulk.

Data Access#

All fine-mapping models and TWAS weights are hosted on Synapse. Access requires a Synapse account and acceptance of the appropriate data use agreement.