TenK10K phase1 multiome project

TenK10K is a population cohort that will profile single-cell RNA-seq data of ~50 million peripheral blood mononuclear cells (PBMCs) from 10,000 individuals in Australia. This cohort will collect

• 5,000 healthy donors, an extended version of the OneK1K cohort
• 2,000 individuals undergoing clinically indicated CT coronary angiogram, the BioHeart cohort
• 1,500 samples from a pan-cancer cohort, the LBIO study (500 donors, 1~5 timepoints for each)
• 1,000 individuals from a pan-autoimmune disease cohort, the AIM study
• 600 samples from a long-COVID cohort, the ADAPT study

This repository will contain the analysis code for computational and statistical analyses of TenK10K phase 1 ATAC/multiome data, with a focus on caQTL (i.e., SNPs associated with chromatin accessibility levels) and multi-omics integration.

In the phase 1 multiome project, we processed 952 TOB individuals for scATAC-seq, 64 BioHeart individuals, and 26 LIBIO individuals for muliome data.

Main data for this project

We obtained PBMCs of 1042 individuals from three sets of scATAC/Multiome data from TOB, BioHeart and LBIO cohorts.

TOB: 952 donors, 119 pools (238 libraries) and 8 individuals per pool (generated in 2024)

BioHeart: 64 donors, 4 pools and 16 individuals per pool (generated in 2022)

LBIO: 26 donors, 4 pools and 6-8 individuals per pool (generated in 2023)

After quality controls, 922 TOB donors were used for caQTL mapping and 60 donors (BioHeart + LBIO) were used for replication.

Preprocessing

• Reads alignment (cell ranger ARC)
• Ambient RNA detection
• Demultiplexing and doublet detection
• Cell type annotation
• Batch correction
• Multi-omics layers integration
• QC & Normalisation

ATAC-seq processing

• Peak calling by MACS2
• DNA accessibility process by latent semantic indexing (LSI)
• Create a gene activity matrix
• Clustering using Azimuth/label transferring
• Verify with scRNA-seq data
• Generate Peak Count matrix

caQTL mapping

• Generate pseudobulk matrix by summing up the ATAC count within each donor
• Correct GC content
• Convert corrected count data to CPM values and normalize the matrix per peak
• Merge the two repeats and estimate principal components
• Perform caQTL with TensorQTL per cell type, fitting covariates

Colocalization

Causal inference

Fine-mapping causal variants

Cell state-dependent effects

Gene regulatory network inference

• Aggregated scRNA-seq and scATAC-seq data per cell type and preprocessed following GLUE’s recommended pipeline
• Constructed a baseline model linking ATAC peaks to genes based on genomic proximity (±150 kb) and eQTL evidence
• Applied GLUE to integrate multi-omics data and infer cis-regulatory peak–gene interactions.
• Built cell type–specific gene regulatory networks (GRNs)
• Incorporated caQTL–eQTL colocalization and SMR results to refine regulatory links and recover additional TF–target relationships.
• Compared cis-regulatory scores between paired and unpaired multiome datasets, integrating GTEx and cell type–specific eQTLs in model training

Data availability

We are currently preparing the data sharing, including full caQTL summary statistics (both common and rare variants), fine-mapping, coloc, SMR, peak-gene links, and GRN inference results. The Zenodo link will be posted here around mid-March 2026.

Citation

Xue et al. Genetic regulation of cell type–specific chromatin accessibility shapes immune function and disease risk. medRxiv. 2025.

Contact

Please send all enquiries regarding this study to the corresponding authors: Angli Xue (a.xue@garvan.org.au) and Joseph Powell (j.powell@garvan.org.au)