How To Use Biological Replicates For Allele Specific Analysis Using Heterozygous Population ?

Question

[Ashutosh Pandey]

Hello everyone,

I am conducting an allele specific expression analysis that consists of five biological replicates. I have RNA-seq data for five F1 hybrid mice. I have calculated allelic ratios (# of reads from parent 1/ # of reads from parent 2) across all the markers for all five biological replicates. I am not sure how to proceed further. I looked at different paper and all of them merged their data from all biological replicates for a marker. The merging of the biological replicates was supported by high correlation among allelic ratios for a marker across different replicates. My replicates too show decent correlations for allelic ratios but are not good as other studies. The only reason I can think of large biological variation is that I am analyzing hippocampus which has heterogeneous population of cells. I can now move further using one of the two methods explained below:

1) For every marker, I can compare the allelic ratio across all the replicates and can remove the outliers (replicates with discrepant allelic ratios) before merging the reads. This will remove discrepant replicate for that marker and merging will increase the number of reads for that marker and that will let me perform statistical tests for allele specific expression with high confidence and calculate p-value.

2) Other option would be to perform separate statistical test and calculate p-value for all the replicates. So for every marker I will end up with 5 p-values (assuming that each replicate has enough reads for a marker so that a statistical test can be carried on) . Then I will have to either combine these p-values or come up with an arbitrary rule. For example if 3 replicates show significant p-value , then the marker shows allele specific expression.

Which approach sounds better to you guys ? Any other suggestion is welcome too. Thanks.

[GouthamAtla]

how did you do it finally ? any updates ?

Answer 1

[Charles Warden]

I am also interested to see how people respond.

Once you identify which alleles correspond to one another, I would assume it would be essentially like a normal RNA-Seq expression experiment (using RPKM expression in limma, DESeq for read counts, etc.). For this type of analysis, it is very important to take biological replicates into consideration:

http://bioinformatics.oxfordjournals.org/content/early/2013/12/06/bioinformatics.btt688.abstract

http://genomebiology.com/2013/14/9/R95

Is the goal to ask if expression varies between the paternal and material alleles? If so, perhaps you can segregate reads into 2 categories: paternal and maternal for each sample. You can then act like you have 10 samples (instead of five) and compare expression between the material subsets and paternial subsets (and you can take the individual mouse into consideration with something like a 2-way ANOVA). However, material and paternal alleles need not actually be the same allele in all F1 mice, so it seems to me like this is not sufficient to address your question.

Maybe my main problem is understanding exactly what biological question you are trying to ask?