gaow
commented
1 year ago Hi @caydenpenn
conditional eQTL mapping with QTLtools,
Could you clarify what you meant by conditional eQTL?
whether I need to include all variants (such as all variants surrounding 1M bp of genes with an eQTL), or just the variants showing signatures in the conditional eQTL mapping analysis
You need to include all variants (after some MAF filter if you like)
Can I simply run susieR for all genes and use the results downstream?
Unless your sample size is huge, for eQTL studies with a few thousand individals (or when the number of individuals is not more than the number of variants to analyze) I would run SuSiE with individual level data directly, although you need to deal with covariates properly #82 . Alternatively, you can also run susie_rss with in sample LD, effect size, standard error and smapel size which is the sufficient statistics set of input and should give you identical result of individal level data fine-mapping
However, I am unsure how to narrow down those candidates further. For example, should I set a PIP threshold greater than 0.8?
It depends on your goal of the analysis. SuSiE credible set (CS) by definition is meant to capture a single effect ie it should contain with eg 95% probability the non-zero effect. But it is hard to tell which one it is. Of course the one with largest PIP inside the CS would be the most likely, and if this PIP is large enough eg >0.8 you will be somewhat confident to claim it is the non-zero effect SNP according to your data. However imagine you have 10 SNPs in complete LD within a CS and one of them is the causal SNP, then all your PIP should be 1/10 = 0.1 because your variants are identical. In this case, it would be hard to select the one SNP from this CS, although it should by design still captures the non-zero effect variant of interest.
caydenpenn
stephens999
pcarbo
kauralasoo
Hello! I am currently working on eQTL fine-mapping analysis with susieR. I have already mapped eQTLs and performed conditional eQTL mapping with QTLtools, which provided me the summary statistics data. I also have individual genotype data. My next step is to perform fine-mapping for these eQTLs. I have several questions regarding this process.
My first question is whether I need to include all variants (such as all variants surrounding 1M bp of genes with an eQTL), or just the variants showing signatures in the conditional eQTL mapping analysis (~200 variants for each gene) with QTLtools. If I need to include all variants, it seems that I did not need to run QTLtools. Can I simply run susieR for all genes and use the results downstream? It is confusing to me whether non-fine-mapping tools are still necessary when individual genotype data is available, given that susieR is both fast and robust at identifying causal variants.
My second question is whether I need to use the summary statistics of the eQTLs (using susier_rss) or just the individual genotype data (using susier) when fine-mapping only the conditional eQTLs from QTLtools.
Lastly, I am wondering if you have suggested cutoffs for extracting causal variants from the susieR results. I plan to use the variants in the 95% credible sets as my causal candidates. However, I am unsure how to narrow down those candidates further. For example, should I set a PIP threshold greater than 0.8? In some credible sets with multiple variants, no single variant exceeds the PIP 0.8 cutoff. How should I interpret such credible sets or variants in those sets? Any guidance would be greatly appreciated. Thank you!