how to evaluate SNPs that are regulating same gene expression across multiple tissues

Hello everyone,

I have a gtex model file, I noticed that there are several SNPs that are regulation Gene A expression in more than one tissues. Is there any method to evaluate such a condition and get function information from it?

This is how my file look like:

enter image description here

So for this:

SNP1(rs1041770) and SNP2(rs12628452) regulating gene ENSG00000283633 is present in tissue adipose subcutaneous but not in adipose visceral.

I was reading research and there is a research being done on eQTL using bipartite network:

This is what the paper states: For each of the 13 tissues, we represented the significant eQTL as a bipartite network, with nodes representing either SNPs or genes and edges representing significant SNP–gene associations

But I want to do it for all tissues, So I have develop network for each tissue: how can I see the commonality and differences between them. Can anyone provide more guidance on this. I am new to network analysis and functional annotation work. Or any other method that can evaluate this kind of relationships for snps across tissues.

Thank you.

snp gtex

• 1,245 views • link

updated 2.9 years ago by LChart

• written 2.9 years ago by rheab1230 •

1 answer

There's a rather significant difference between the "network structure of eQTL" analysis (CONDOR, etc.), and performing functional annotation. The latter is ideally a bit more strict, and really needs to incorporate fine mapping. Specifically, for a single tissue, an "edge" would reflect whether a SNP is in the credible set for the expression of a gene (see https://www.nature.com/articles/s41467-021-23134-8#Sec9).

It seems like you're interested in tissue specific effects; in which case the fine-mapping model should be updated to be of the form

expression ~ tissue + genotype_i + tissue:genotype_i

and ask for whether the SNP is in the credible set for the cross-term. Note that you do have to be careful; as a variant which has a strong effect in all tissues except one will look like a specific-effect (in the opposite direction) for that tissue.

Note that in the limit of infinite data, every tissue will necessarily have a tissue-specific term, if only due to pure cell compositional effects, so a general methodology probably won't answer particular questions about tissue-specific action of regulatory variants (which typically would take the form of a contrast, e.g., "stronger in intestine than stomach").

That said, a good entry-point for bipartite network analysis might be https://europepmc.org/article/pmc/6333914. Bipartite networks occur regularly in:

miRNA targets: (miRNA, RNA) edges

drug targets: (drug, predicted binding target) edges

receptor-ligand interactions: (receptor, ligand) edges

TF-gene networks: (TF, regulated gene) edges

so literature in all of these areas could provide inspiration for the analysis you might want to perform.

• 0 views • link

written 2.9 years ago by LChart

I should add that for scoring genomes (i.e., from sequencing data) you do not need this kind of fine mapping; and can use pre-computed expression weights like PrediXScan (https://predictdb.org/post/2021/07/21/gtex-v8-models-on-eqtl-and-sqtl/) which uses elastic net to select the variants. There will be many variants in the "credible set" with weight 0 due to the elastic net -- which is why you can't really compare weights across tissues -- but since they're all tagged by the selected variant, the prediction does not suffer.

You can use the scores for multiple tissues in multiXscan (https://journals.plos.org/plosgenetics/article?id=10.1371/journal.pgen.1007889) for gene association; and potentially analyze the model results to contrast the effects of different predicted tissue expression.

• 0 views • link

written 2.9 years ago by LChart