Correlate expression levels with phenotypic measure?

[glados]

Dear all.
I've been spending lots of time searching for an answer to this question, going through papers and discussing with people.

We have gene expression levels in 3 individuals that we want to correlate with a phenotypic measure, lets say levels of Hg in the blood. How should we best go about thinking about this? We have tried running pearson correlations on expression levels vs Hg levels and then only using the top and bottom 1% correlated genes. We have tried to randomly sample Hg levels from a pool of 6 sequenced individuals with different Hg levels to run pearson's correlations 1000 times and later select only the genes that have real correlations outside the 95% confidence interval. One problem is that we're doing correlations with only three data points and that will give lots of false positives.

Any tips and suggestions on how to proceed?

[TiborNagy]

Try to make two Hg level class (for example high and low) and make a differential expression analysis.

[sindrle]

Im not good at this, but what happens in you input Hg as a continous (?) variabel into edgeR or DEseq2 as a co-variat?