haQTLQTL resultsROSMAPDLPFC
ROSMAP DLPFC H3K9ac QTL
Religious Orders Study (ROS) or the Rush Memory and Aging Project (MAP) h3k9ac QTL analysis using the FGC xQTL pipeline.
Please refer to this document for an overview of the ROSMAP project.
Get the data
Controlled access. Request data access for ROSMAP on the AD Knowledge Portal; you need a free account there. See How to get the data.
| What you need | Open, no application NIAGADS NG00184.v1 | Controlled access AD Knowledge Portal |
|---|---|---|
| Association summary statistics | all · bh_significant · hmt_significantDownload from NIAGADS ↗ | Use the open release |
| Fine-mapping | single_context_finemapping_cs95 · single_context_finemapping_allDownload from NIAGADS ↗ | Use the open release |
| Look up a gene, variant or region | xQTL Atlas ↗ | Not needed |
A AD Knowledge Portal ID opens a folder; synapse get -r with that ID downloads everything in it. Open files are packaged by data type, so one NIAGADS archive holds every haQTL dataset in that release.
pip install synapseclient
synapse login
synapse get -r syn69670592Every AD Knowledge Portal folder that mentions this dataset 11 locations
Variant and gene summaries5
- xQTL_finemapping › single_context_finemapping › haQTL · syn69865749
ROSMAP_DLPFC_haQTL.cis.top_loci.bed.gz
- xQTL_loci_summary › xQTL_finemapping › multi_context_finemapping · syn69865754Covers all ROSMAP data types; also listed on the cohort's other QTL pages
ROSMAP.cis.multi_context.top_loci.bed.gz
- xQTL_loci_summary › xQTL_colocalization › multi_context_colocalization · syn69865796Covers all ROSMAP data types; also listed on the cohort's other QTL pages
xqtl_only_ROSMAP_colocboost_sorted.bed.gz
- xQTL_loci_summary › xQTL_colocalization › trans_colocalization · syn77579916Covers all ROSMAP data types; also listed on the cohort's other QTL pages
xqtl_only_ROSMAP_transmap_colocboost_sorted.bed.gz
- AD_loci_summary › AD_xQTL_colocalization · syn69865816Covers all ROSMAP data types; also listed on the cohort's other QTL pages
AD_xQTL_ROSMAP_colocboost_sorted.bedAD_xQTL_ROSMAP_transmap_colocboost_sorted.bed.gzAD_xQTL_ROSMAP_ISSAC_colocboost_sorted.bed.gz
xQTL and AD GWAS models3
- xQTL_models › colocalization_modelsCovers all ROSMAP data types; also listed on the cohort's other QTL pages
ROSMAP syn75362558
batch{1..2} syn75427341+1 batch
ROSMAP.*.cb_xqtl.rds
- xQTL_models › TWAS_models › multi_contextCovers all ROSMAP data types; also listed on the cohort's other QTL pages
ROSMAP_All syn74909252 FGMB Atlas of multi-context RWAS prediction models
batch{1..2} syn75427331+1 batch
ROSMAP_All.*.multicontext_twas_weights.rds
- AD_xQTL_integration › AD_xQTL_colocalization_modelsCovers all ROSMAP data types; also listed on the cohort's other QTL pages
ROSMAP syn74986543
ROSMAP.*.cb_xqtl_AD_Bellenguez.rdsROSMAP.*.cb_xqtl_AD_Bellenguez_EADB_2022.rdsROSMAP.*.cb_xqtl_AD_Kunkle_Stage1_2019.rdsROSMAP.*.cb_xqtl_AD_Wightman_Excluding23andMe_2021.rdsROSMAP.*.cb_xqtl_AD_Wightman_ExcludingUKBand23andME_2021.rdsROSMAP.*.cb_xqtl_AD_Wightman_Full_2021.rds
Reference files3
- xQTL_ldscores › single_context › haQTL
ROSMAP_DLPFC syn75185960
ROSMAP_haQTL_DLPFC_single_1.{1..22}.l2.MROSMAP_haQTL_DLPFC_single_1.{1..22}.l2.ldscore.parquetROSMAP_haQTL_DLPFC_single_1.{1..22}.l2.ldscore.parquet.logROSMAP_haQTL_DLPFC_single_1.{1..22}.annot.gz
- without_hm3_print_snp › xQTL_ldscores › colocalizationCovers all ROSMAP data types; also listed on the cohort's other QTL pages
ROSMAP syn75186616
ROSMAP_colocboost_single_1.{1..22}.l2.MROSMAP_colocboost_single_1.{1..22}.l2.ldscore.parquetROSMAP_colocboost_single_1.{1..22}.l2.ldscore.parquet.logROSMAP_colocboost_single_1.{1..22}.annot.gz
- without_hm3_print_snp › xQTL_ldscores › multi_contextCovers all ROSMAP data types; also listed on the cohort's other QTL pages
ROSMAP syn75187008
ROSMAP_Multi_All_single_1.{1..22}.l2.MROSMAP_Multi_All_single_1.{1..22}.l2.ldscore.parquetROSMAP_Multi_All_single_1.{1..22}.l2.ldscore.parquet.logROSMAP_Multi_All_single_1.{1..22}.annot.gz
Folders and files whose names carry this dataset's cohort, data type and tissue (bulk and single-nucleus data kept apart), taken from the staging folder list. All AD Knowledge Portal folders lists everything.
Contact
Xuanhe Chen
Study Overview
- Study name : ROSMAP DLPFC H3K9ac QTL
- Study Description : Religious Orders Study (ROS) or the Rush Memory and Aging Project (MAP) h3k9ac QTL analysis summary statistics using the FGC xQTL pipeline.
Samples’ phenotype information (sex, age, race etc.) can be found in ROSMAP metadata, see “Other information” section in this document
Analysis Status
TransQTL association: Need to be performed.
Analysis Details
Sample size tracking through the analysis workflow
| Step | Sample Size |
|---|---|
| Raw data | 669 |
| Sample 90214403 as been removed from 2022Feb08 ROSMAP | 668 |
| After overlapping with ROSMAP WGS | 607 |
| After overlapping with ROSMAP joint call | 596 |
QC and Normalization Details for Phenotype Data
Used h3k9ac readcount data directly without further QC. For QCs performed in Chip-seq and peak calling please refer to descriptor/omics/histone_ChIPSeq/ROSMAP_DLPFC_ChIPSeq_H3K9ac.md
Genotype parameters
- Kinship coefficient for related individuals: 0.0625
- Minor allele count cutoff for single variant analysis: 5
- Variant level missingness threshold: 0.1
- Sample level missingness threshold: 0.1
Covariate parameters
PCA analysis:
- MAF for PCA: 0.05
- LD pruning via PLINK for PCA analysis: window 50, shift 10, r2 0.1
- Tolerance of missingness in covariates: 0.3
- number of pc that captured more than 70% PVE were kept
PEER analysis (Default values from PEER software):
- The number of max iteration: 1000
- Prior parameter Alpha_a: 0.001
- Prior parameter Alpha_b: 0.1
- Prior parameter Eps_a: 0.1
- Prior parameter Eps_b: 10.0
- Tolerance parameter tol: 0.001
- Tolerance parameter var_tol: 0.00001
- minimum variance explained criteria to drop factors while training r2_tol: False
- Convergence mode: “fast”
Final covariates: Sex, Age at death, PMI, PCs(13), PEER factors(60)
TensorQTL association scan parameters
- cis window for the up and downstream radius to analyze around the region of interest, in units of bp: 1000000
- Minor allele count cutoff: 5
- Minor allele frequency cutoff: 5/(2N), where N is the number of samples in the analysis. The pipeline computes N from the covariate file, so the cutoff differs between datasets.
Links to QTL analysis notebooks
See notebooks in:
The notebooks in this folder contain the commands and data wrangling codes for analysis of the h3k9ac data in ROSMAP. (data wrangling exist because not all data are processed using the xqtl-pipeline from the beginning and need to be reformatted to fit one intermediate step of the pipeline).
Association data preprocessing
Genotype data preprocessing
We conducted Whole Genome Sequencing (WGS) on 1160 DLPFC samples as part of the ROSMAP study. Post-sequencing, these samples were aligned to the GRCh38 reference genome, and the resulting genetic variants were cataloged and saved in VCF (Variant Call Format) files. To ensure the integrity of our findings, we applied quality control. This involved the utilization of the GRCh38 reference genome complemented by the GENCODE 34 annotation. Our quality control also used the ERCC spike-in annotations. The parameters were: a read depth (–DP) of 1, an indel depth (–DP-indel) of 1, and a Genotype Quality (GQ) threshold exceeding 20. Furthermore, each site was mandated to have at least one sample surpassing the allele balance threshold, specifically set at >= 0.15 for Single Nucleotide Polymorphisms (SNPs) and >= 0.20 for insertion-deletion mutations (indels). After the initial QC, the VCF files were merged using PLINK for downstream analysis. Subsequently, we applied further filters to exclude variants with more than 10% missing data and set a Hardy-Weinberg Equilibrium (HWE) cutoff of 1E-8. The generated genotype files will be subjected to subsequent Principal Component analysis (PCA) as part of the covariates, as well as utilized for QTL mapping.
Principal component analysis for haQTL mapping (this would be the new pipeline)
For the generation of a covariate file for QTL mapping, we initiated the process with Principal Component analysis (PCA) applied to both genome and phenotype data, incorporating several pivotal steps. In the quality control stage, we utilized the KING to segregate samples into related and unrelated categories. PCA was then performed on the unrelated samples using the flashpcaR library, followed by outlier detection to identify pertinent genotype PCs. We retained those with a cumulative explained variance (PVE_cum) of 0.8 or higher and combined them with relevant metadata sexm, gender and post-mortem interval (pmi). Subsequently, we calculated residuals on these merged covariates and embarked on hidden factor analysis to identify potential batch effects or artifacts in the phenotype data. The number of principal components to be included was determined through a comparison of their eigenvalues to the Marchenko-Pastur Distribution. The principal components identified as hidden factors were incorporated with genotype PCs and metadata into the final amalgamation, serving as the basis for subsequent QTL mapping analysis.
phenotype preprocessing can be run in parallel but covariate preprocessing needsto be done last since it requires both phenotype data and PCs from PCA in genotype preprocessing.
- phenotype_calling shows the commands used for phenotype calling.
- genotype_preprocessing.ipynb lab access only shows the commands used for genotype processing and preparation steps.
- Includes QC, PCA, and PCA result figures, for example h3k9ac_whole.k9_cov.xqtl_protocol_data.filtered.h3k9ac.related.filtered.extracted.pca.projected.resid.PEER.diag.pdf lab access only is the view of PCA result of the haQTL analysis.
- phenotype_preprocessing.ipynb lab access only shows the commands used for the phenotype data processing and preparation steps.
- This notebook strated from some already cleaned data and convert them to bed format inputs taken by tensorQTL.
- covariate_preprocessing.ipynb lab access only shows the commands used for the covariate data processing and preparation steps.
- This includes PEER factor analysis steps and merging the PEER factors to covariates in raw data and PCA.
Association scan using TensorQTL and summary statistics standardization
For molecular cis-QTL mapping, we utilized quantified H3K9ac levels in BED format as phenotypes, principal components (PCs) obtained as previously described as covariates, and corresponding genotype data from the same dataset in PLINK format, as delineated in prior descriptions. We employed the TensorQTL pipeline to perform the association analysis, setting a minor allele count (MAC) cutoff of 5. This approach allowed us to generate phenotype-level summary statistics with empirical p-values, facilitating the calculation of genome-wide false discovery rates (FDR) through permutations. Additionally, summary statistics were obtained for all variant-phenotype pairs. During this process, we expanded the cis window to capture signals that were not confined to the coordinate level, but rather operated on the spatial level within a specific region. The cis window was defined as a region encompassing 1,000,000 base pairs, in addition to the topologically associated domain (TAD) and its boundary (TAB). This was executed with an understanding that variants located within this spatial domain could exert an impact on gene expression, thus associating with the corresponding gene. In instances where a gene was positioned within the confines of two generalized TABD boundaries, we systematically selected the outermost TABD boundary for incorporation into the cis window definition.
SuSiE univariate fine-mapping
- univariate_fine_mapping.ipynb contains the workflow of haQTL univariate fine-mapping.
Dataset Details
- Number of cases :
- Number of controls :
- Annotation information :
- Genotype level imputation :
- Imputation Panel :
Path(s) to genotype matrix
output of genotype_preprocessing.ipynb
Path(s) to omics-data matrix
output of phenotype_preprocessing.ipynb
Path(s) to covariate data matrix
output of covariate_preprocessing.ipynb
Path(s) to cis-QTL association testing
output of cis_haqtl_association.ipynb
/projects/histone-methylation/CU/h3k9_sumstats/
Path(s) to fine-mapping with SuSiE model
output of univariate_fine_mapping.ipynb
Path(s) to fine-mapping with fSuSiE / SuSiE RSS model
Fine-mapping model objects (fSuSiE/SuSiE-RSS, .rds) are available in the fine-mapping models folder (syn69670592).
Path(s) to colocalization
Pairwise colocalization models (SuSiE-coloc): syn69670597
AD GWAS–xQTL colocalization results: syn69865816
AD GWAS–xQTL colocalization models: syn69670630