Gene expression level from replicates

[gen2prot]

Hello All,

I have RPKM values for two treatment conditions (lets say control and diseased). Each treatment has 3 replicates. I have transformed all the replicate data like so: [log2(count+1)]. Now if I wanted to get a single value of gene expression for control and diseased, what do I need to do? Average, Median or sum of replicates.

I read a number of papers where data transformation is discussed (log2(count+1)) but could not find references on how to obtain a single value quantifying the gene expression in control and diseased.

The ultimate goal of course is differential gene expression. For this, is the above at all necessary?

Any help will be very beneficial.

Thanks

[dpryan]

Is there any particular reason that you're not using one of the many R packages or stand-alone programs oriented toward doing this all for you?

[gen2prot]

No particular reason. I was curious that if one were to consolidate the gene expression values from the replicates, and arrive at a single value, how would one do it. Average, median, sum any other. I'd go with median, but was wondering if there was a consensus or not. I could not find any in public literature.

I know there are lots of methods for differential gene expression in R (DESeq, DEXSeq, EdgeR). But for quantifying and arriving at a single value I am not sure. Please advice.

Thanks

[dietmar13]

as especially cancer is (usually) a very heterogeneous disease none of your ideas is very convincing. mean (and similar sum) will give outliers a too big weight and the median will hide outliers and values far away from the median.

i think, gene expression analysis on biological replicates is usually used for showing the variance of gene expression and using this information to assess significant differences.

however, if you are interested in a "typical" gene expression height of your conditions you should use the median (of more than three replicates) or something like the trimmed mean.

the median of
1
2
100
is 2, but do you think this is the answer you want?

the mean is 34.33 (?)

the trimmed mean does not work at all with only three replicates. if you trimm > 33.33% you get 2 (the median) or < 33.33% you get the mean.

[gen2prot]

I agree RNA-Seq gene expression analysis is mostly used for DE estimation. But it may also be used to assess the level of expression (high, med or low just to give an example) w.r.t reference set of genes for example UBI or ADH. In that case having a single value helps, and I think median would largely make sense. Sum biases the data towards the maximum and average is affected by outliers.

Thanks for the input. I think I have my answer.