PrediXcan Example Results

142 views
Skip to first unread message

Elexis Allen

unread,
Mar 7, 2021, 4:34:05 PM3/7/21
to PrediXcan/MetaXcan
Hello!

I recently ran the Example provided for PrediXcan titled "Example for Prediction and Association" (located towards the bottom of the linked webpage). The output was two text files related to predicted expression and association. I wanted a little clarity on the meaning of each column/row, as well as the values associated. 

The association file has 6 column titles: gene, beta, t, p, and se(beta). I'm aware of the significance of the gene column, but the other five are what I'm not so sure of. What does each mean and what do the values associated with each column mean? Anything specifically we should look for that hints at association amounts? 

The predicted expression file has two columns, FID and IID. I believe that FID is associated with different genes. What is IID and what is its significance? Anything to look out for when relating the values associated with the IID column to the predicted expression of each gene? 

Thank you!

Best, 

Elexis


nsanthanam1

unread,
Mar 8, 2021, 7:22:51 PM3/8/21
to PrediXcan/MetaXcan
Hi Elexis,
Just a brief summary of how PrediXcan works. The first output is the predicted expression file; using the genotypes provided, PrediXcan will predict expression of a number of genes. The second output file is the association file which runs an association between the predicted expression from the previous file and a number of phenotypes provided. 

In the association file, the beta refers to the coefficient of the fit between the predicted expression of the gene and the phenotype of interest. The SE is the standard error of that beta. Together the beta value and its error give you an idea of the effect size. T is the value of the test statistic when performing an association test between the predicted gene expression and the phenotype. From the test statistic, the p column represents the p-value. A small p-value would indicate that there is a significant association between that gene and phenotype. From your results, ideally you would want a p-value that is significant at the Bonferroni threshold (0.5/#genes) or (0.5/#genes*#phenotypes), if you have multiple phenotypes you're testing for.

In the association file, each line represents an individual. The IID refers to that person's distinct ID and the FID would be a familial ID. If there are multiple people from the same family, they would have the same FID, and is a measure for relatedness. This file just details what is the phenotype value for each individual. 

Hopefully this all helps! 
Natasha

Alvaro Barbeira

unread,
Mar 8, 2021, 9:58:58 PM3/8/21
to PrediXcan/MetaXcan
Hi there!

Please bear in mind that https://github.com/hakyimlab/PrediXcan is deprecated. The functionality was migrated to https://github.com/hakyimlab/MetaXcan , and some additional features like HDF5 and VCF support were added,along some bug fixes and support for the newer MASHR models.

Best,

Alvaro



--
You received this message because you are subscribed to the Google Groups "PrediXcan/MetaXcan" group.
To unsubscribe from this group and stop receiving emails from it, send an email to predixcanmetax...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/predixcanmetaxcan/638e3317-9d13-485b-8d93-61bae5468b61n%40googlegroups.com.
Reply all
Reply to author
Forward
0 new messages