Meta-analyses to combine significant genes

[inestm28]

I was looking into your code at https://github.com/dhimmel/stargeo/blob/master/combine.ipynb.

Could you tell me where the file 'balanced_permutation.tsv.gz' comes from /is downloaded from?

And just to make sure, this code uses only the p-values in the "multipletests" method, right?

So, it first gets all the significant genes from each independent study, then creates corrected p-values and after selects the genes that have "up" direction through the fold-change? I'm not sure how the process goes. I would really appreciate if you could help me, please.

And I read in this paper https://doi.org/10.1093/bib/bbaa019, that random effects models are used when the gene is significant in all studies. So it is not used for when it's not in some of the studies, right?

[dhimmel]

Hey @inestm28. Thanks for your questions. I don't remember much about this analysis, so the source code is probably the best reference for the methods. Here are some additional places to look.

• Rephetio Manuscript methods
• STARGEO manuscript
• Thinklab discussion from Rephetio

Could you tell me where the file 'balanced_permutation.tsv.gz' comes from /is downloaded from?

The querier.ipynb notebook creates balanced_permutation.tsv.gz for each disease.

So, it first gets all the significant genes from each independent study, then creates corrected p-values and after selects the genes that have "up" direction through the fold-change?

The combine.ipynb notebook appears to use the random_pval_corrected and renames it to p_adjusted, which gets written to data/diffex.tsv.

Sorry if these answers don't address everything. Feel free to continue using this issue to leave your notes and conclusions for these questions.