about missing file PDB_2374_MoBa_Q1_Q4_MBRN_full.RData

[mugpeng]

Hi,

Did the file ~/archive/START/pheno/PDB_2374_MoBa_Q1_Q4_MBRN_full.RData open for others for test?
Cuz I didn't find it in this project.

Thanks.

[mugpeng]

Or would you mind to explain the process that you handled the gwas instead of stimulation data for CFMR? thanks.

[william-denault]

Hello Peng,

Thank you for your interest in our work. Unfortunately the file ~/archive/START/pheno/PDB_2374_MoBa_Q1_Q4_MBRN_full.RData is not publicly available. Unfortunately, I haven't had time to write a complete software for CFMR. I am currently busy. I will come back to you in a day or two with a detailed explanation of the pipeline that should allow you to adapt it easily.

Thank you very much for your interest and patience.
Best
William Denault

[mugpeng]

Hello Denault,

Thanks for your reply. And I am looking forward to your new adaption.

[william-denault]

Hello Peng,

Thank you for your interest in our work. Unfortunately the file ~/archive/START/pheno/PDB_2374_MoBa_Q1_Q4_MBRN_full.RData is not publicly available. Unfortunately, I haven't had time to write a complete software for CFMR. However, as you may have understood, a complete pipeline is available in the folder /CFMR/CFMR_pipeline_code.

The pipeline should be easy to adapt for your data. Please find a description of different scripts below:

First, you need to define a Data folder where we will store all the files that are generated.
Let's say you want to perform a 10-fold CFMR. You will need to generate the following additional subfolders:

Data/Split_1_GWAS_Maternal
Data/Split_2_GWAS_Maternal
.....
Data/Split_10_GWAS_Maternal

and the folders:

Data/Extracted_SNPs_CHR1
Data/Extracted_SNPs_CHR2
...
Data/Extracted_SNPs_CHR10

(you can, of course, change Split_k_GWAS_Maternal, by, for example, Split_k_GWAS)
These folders will store the output of the 10 different GWASes that are performed in the script 2_Running_GWAS.R)

Script 3_Clumping.R will clean the output of the 10 GWAS (basically removing markers with a MAF below 5%, you see that at the top of the script). The cleaned files are named "Split_k_GWAS_mother.txt" in the folder Data/Split_k_GWAS_Maternal. Then it will clump the results of each GWAS only keeping SNP with a value below 10^-2 (you can, of course, use your own clumping setting, in this script, we use something quite basic). The clumped results are stored in the file Split_k_SNP_for_extraction.txt, again saved in the same folder Data/Split_k_GWAS_Maternal.

Script 4_Extracting_SNPs.R will basically extract the SNP previously clumped and write them into the folder Data/Extracted_SNPs_CHRk

Script 5_buidling_IV.R may need more adaption. Please let me know when you are there. We would need to adapt the IVs_build function to your data. But basically, it builds a different type of instrument depending on the threshold selection for Cross fitted instrument.

Once we have passed script 5 then the last steps are straightforward.

Looking forward to hearing from you,
Best regards,
W

[mugpeng]

Cooool! Very much thanks for your selfless and quick help. When I finished script 5 will let you know. And thanks again. :)

[mugpeng]

Hello Denault,

It's very nice of you for helping me, and I successfully complete this pipeline in my work.
But I still have one question.

In the simulation data which exhibiting in your readme, we know the effect of X on Y is 1(it's predefined), so it's easy to compare each methods. But in read data, we don't have prior knowledge about the true effect, how can we judge each method?
I've seen some methods also mentioned and included in ivreg package(diagnosis approach). Whereas I don't think they can be used to evaluate methods if they predict the result accurately.

Thanks.

[william-denault]

Dear Peng,

Sorry for the delayed answer.

Please let me know if you need further help with your analysis (you should not be too far from the end of it now), regarding your point. It is true that the readme only includes to comparison with other ivreg routines. If you wish, I can point you to the other script used for the simulation in the manuscript. Regarding your point, I am not sure which part you refer to exactly:

It is correct that the true of X on Y is not known in general, and it is still an open question how to be certain of such an effect ( see https://academic.oup.com/ije/article/45/6/1866/2930550 for a thoughtful discussion on this matter). MR is only one way to get an idea of this effect. You can use different MR approaches (each one of them has assumptions, as well as pro and cons) and see if you observe some consistencies across this method (MR Egger, MR RAPS, CFMR, CAUSE). Furthermore, maybe in your setting one of these methods is more likely to have its assumption fulfilled (and therefore should be trusted more than some other approaches). You can also have a look there to strengthen your analysis https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5133381/

Hopefully, you can conduct different analyses (different potential papers) using different approaches and observe some consistency. You can have a look to this paper https://www.nature.com/articles/s41598-021-85684-7 from a solid group in MR.

I hope this is useful. Let me know if I should help you with the next steps.

Best
W

[mugpeng]

Dear Denault,

Thanks a lot for your very patient reply. I will read the material mentioned above.
In my point, I am trying to find some indicators that can be used to evaluate methods if they predict the result accurately, like CFMR, basic one-sample MR……