eQTLs: how much are they tissue specific vs shared?

[stephens999]

Put together an rmd file, with plots, showing main features of eQTLs identified. Eg, are they tissue specific, or shared. Which tissues tend to share eQTLs?

Possible plots/analyses include:

• histogram of some measure of heterogeneity/homogeneity of tissue effects. eg could be "number of tissues for which effect exceeds some threshold" or some kind of variance of effect sizes (or coefficient of variation of effect sizes etc). The idea is that this histogram would give an overall impression of "tissue-specificity". My expectation is that it might even be bimodal - that is, some eQTLs will be shared among all or many tissues, whereas others will be in a very small number of tissues. There is a similar plot in showing this "U" shape in the gtex pilot paper.
• along similar lines, some kind of clustering of eQTLs aimed at clustering the "tissue-specific" eQTLs together, and the "tissue-constant" eQTLs together. This might require thought about the right metric for clustering. Eg you could first sort the effect sizes before clustering, so that the tissue-specific ones look something like (0,0...,1) and the tissue constant ones look more like (1/k, ... ,1/k).

Possibly would be useful to apply these analyses to lfsr instead of posterior means? So in this case first plot would be somethign like a histogram of "number of tissues showing lfsr<0.05" for each eQTL.

Given the fact that brain tissues look somewhat different, might be worth doing the above separately for brain and non-brain tissues.

[surbut]

Here are some examples of tissue specific and fully shared eQTLS as quantified by posterior (shrunken) t stats using my Matrix Ash on 1188 components (ID, 5 SFA, rank 2 PC, rank 5 SFA, and 45 singleton/full configs) https://github.com/stephens999/gtex-viz/blob/master/Plots/tissuespec.Rmd and here is the distribution of the number of tissues with lfsr<0.10 per gene. https://github.com/stephens999/gtex-viz/blob/master/Plots/lfsrhist.Rmd

[stephens999]

thanks sarah. Given you've pushed the html files too, probably best to link to those.

[stephens999]

@surbut: can you prepare "nice" versions (ie with reader-friendly axis labels and titles like "number of tissues in which eQTL is "significant"") of a) the histogram, b) the t stats for a tissue-specific testes example; c) the tstats for a consistent example?

[surbut]

Sure - here is the lfsr hist https://github.com/stephens999/gtex-viz/blob/master/Plots/lfsrhist.html

[surbut]

just updated the lfsr histogram - I'm not sure what I should change in the tissue specific and shared examples since those are my 'nice plots' :) I put my plot function in the rmd if you'd like to change it?

On Jun 19, 2015, at 11:07 AM, stephens999 notifications@github.com wrote:

@surbut: can you prepare "nice" versions (ie with reader-friendly axis labels and titles like "number of tissues in which eQTL is "significant"") of a) the histogram, b) the t stats for a tissue-specific testes example; c) the tstats for a consistent example?

— Reply to this email directly or view it on GitHub.

[surbut]

added spacing and full title name to tissue specific shared example https://github.com/stephens999/gtex-viz/blob/master/Plots/tissuespec.html

[surbut]

added proportion of eQTLs shared by two tissues https://raw.githubusercontent.com/stephens999/gtex-viz/master/Plots/sharedQTLs.html

[stephens999]

hmm. unfortunately linking to html doesn't help as I'd hoped since git doesn't render it nicely....

[surbut]

sorry - should I revert to linking to Rmd?

On Jun 19, 2015, at 1:00 PM, stephens999 notifications@github.com wrote:

hmm. unfortunately linking to html doesn't help as I'd hoped since git doesn't render it nicely....

— Reply to this email directly or view it on GitHub.

[surbut]

Here are tissue specific .Rmds:

https://github.com/stephens999/gtex-viz/blob/master/Plots/tissuespec.Rmd

Intersection heatmap:

https://github.com/stephens999/gtex-viz/blob/master/Plots/sharedQTLs.Rmd

lfsr histogram:

https://github.com/stephens999/gtex-viz/blob/master/Plots/lfsrhist.Rmd

[surbut]

added interactive sharing plot: https://github.com/stephens999/gtex-viz/blob/master/Plots/shinytest.Rmd

[surbut]

https://github.com/stephens999/gtex-viz/blob/master/Plots/tissuespecificbytissue.Rmd This is the plot of the number of QTLS that are tissue-specific per tissue. It is a shiny plot

[surbut]

Also added 'unshiny' with the tissue names above https://github.com/stephens999/gtex-viz/blob/master/Plots/tissuespecificbytissuenoshiny.Rmd. Just created a boolean matrix for each snps on whether the tissuelfsr was <=thresh, and then took the row sums. The rows with row sum =1 merited tissue-specific distinction, so we subset those rows, and then sum the columns to get the tissue specific hist.

[surbut]

really cool tissue specific barplot with interactive labels: https://plot.ly/~DemoAccount/866 Here's the rscript for producing it: https://github.com/stephens999/gtex-viz/blob/master/Plots/tissuespecificbarplot.Rmd