questions about MR-JTI

[benruiWu]

hi, thank you for your wonderful method! i have problems when using MR-JTI to interpret my GWAS results.

problem 1: do i need to run JTI methd before runing MR-JTI?
if not, how can i get the eqtl value?

problem 2: how to predict the genes' expression data using JTI after training? (i didn't find "predict" function in README)

problem 3: how to interpret the MR-JTI result table? dose the "sig" mean that the gene is a positive result for the test of causality to the trait?

[zdangm]

Hi Benrui, Thanks for using MR-JTI!

1 Technically, the MR-JTI doesn't request JTI results to run. However, we suggested running MR-JTI for genes with significant signals in JTI. The MR-JTI takes the marginal effect size of eQTL. If you have the access to the individual-level data, you could run a variant-gene cis eQTL association test to get the marginal effect size. The marginal effect size is also available from the GTEx portal. However, the platform started to charge for these large eQTL file downloads (via google cloud). As a potential compromise, you may use the marginal effect size estimated from the earlier GTEx version (v7) which is free to download.

v7 https://storage.googleapis.com/gtex_analysis_v7/single_tissue_eqtl_data/GTEx_Analysis_v7_eQTL_all_associations.tar.gz

2

The product of a genotype matrix (sample by SNP matrix, the genotype is coded by 0, 1, and 2) and a vector of weights (for each SNPs. The weights can be accessed from the .db file) is the genetically regulated expression(GReX). In the current pipeline, we assumed that you will use the summary-statistics-based method to estimate the association. So, we didn't include the script to calculate the GReX. Please let me know if you have any additional questions about generating the GReX.

3

Yes. A 'sig' in the first line (started with "expression ... ") suggests a potential causal effect from the gene to the trait.

Dan

[benruiWu]

Thank you very much！ your advices really help！:)

however one more thing still confuses me. i don't know how to clump snps around genes in batches and i don't know what is the criteria for snps to clump too

i am looking forward to your help!

Benrui

[zdangm]

Benrui,

The plink --clump cmd can handle it. If you would like to clump based on the p-value of eQTL signals for each gene, you may perform clumping only for the cis region (--chr --from-bp --to-bp). The suggested threshold is included in the example below.

e.g., plink --bfile ${geno_path} \ --chr ${chr} \ --from-bp ${cis_region_start} \ --to-bp ${cis_region_end} \ --clump ${eqtl_path} \ --clump-p1 1 \ --clump-r2 0.1 \ --out ${output_path}

Let me know if you have any additional questions. Dan

On Fri, Aug 20, 2021 at 7:34 AM benruiWu @.***> wrote:

Thank you very much！ your advices really help！:)

however one more thing still confuses me. i don't know how to clump snps around genes in batches and i don't know what is the criteria for snps to clump too

i am looking forward to your help!

Benrui

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/gamazonlab/MR-JTI/issues/8#issuecomment-902659441, or unsubscribe https://github.com/notifications/unsubscribe-auth/AH5HAMNBXAOMZFMGGHS5WULT5ZDTXANCNFSM5CLR4KCQ . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&utm_campaign=notification-email .

[erskck1]

Hi Dan, Thank you very much for this amazing tool. I would like to use it to generate imputed gene expression data from individual GWAS data ( not summary statistics), like PrediXcan does. Regarding the Benrui's problem 2, I have a similar question. I don't know how Benrui did it but I need some help to generate GReX. Could you give me some advices?

Best wishes, Ersoy

[zdangm]

Hi Ersoy, The GReX can be generated by Gb, where G is a sample x genotype matrix coded by dosage (0, 1, 2) and b is a vector of weight. Please flip the allele if the genotype counted allele (alternative allele, typically A1 in plink format) and the effect allele in the weight file are not inconsistent. If you don't want to generate it manually, the PLINK --score function can be used to generate the GReX (PLINK will handle allele flipping automatically). The weight file can be found in the .db file. The attached lines may help to pull the weights. Dan

library(RSQLite) con <- dbConnect(RSQLite::SQLite(), dbname='xxxx.db') #establish connections dbListTables(con) #datasets dbListFields(con, 'weights') #cols weights = dbReadTable(con,"weights") dbDisconnect(con) #disconnect

On Tue, Jun 14, 2022 at 11:50 PM Ersoy Kocak @.***> wrote:

Hi Dan, Thank you very much for this amazing tool. I would like to use it to generate imputed gene expression data from individual GWAS data ( not summary statistics), like PrediXcan does. Regarding the Benrui's problem 2, I have a similar question. I don't know how Benrui did it but I need some help to generate GReX. Could you give me some advices?

Best wishes, Ersoy

— Reply to this email directly, view it on GitHub https://github.com/gamazonlab/MR-JTI/issues/8#issuecomment-1155385453, or unsubscribe https://github.com/notifications/unsubscribe-auth/AH5HAMPZOOXNYDZO6EAY3TDVPCS3NANCNFSM5CLR4KCQ . You are receiving this because you commented.Message ID: @.***>