It's common to talk about expression similarity with clustering, correlation coefficients, etc. I want to do something slightly different. I want to argue that, biologically, my two samples are very different. I have microarray expression profiles from both. I have shown:
1) Negative Pearson/Spearmann correlations for genome-wide comparisons.
2) The genes differentially expressed in one sample aren't differentially expressed in the other.
This doesn't seem like it's enough. Any suggestions?
Not clear what you mean by "genes differentially expressed in one sample aren't differentially expressed in the other." Differential expression means significantly different expression under one condition as compared with another (e.g. cancer versus normal, drug-treated versus control) - i.e. we compare groups of samples. So what are you comparing to what?
I agree with the comment above and also think it would be easier to help you with more details. What are you samples, conditions,... Can you detail a bit more please?
Can you look at the marks(TFs, histone modification marks, CpGislands etc.) your data contains and then may be compare it say expressed genes with specific marks or expressed vs non-expressed genes with a specific or group of specific marks(active chromatin, repressive etc.)
Transcriptional response: With different sets of genes differentially expressed in response to different treatments, one could look for common regulators of the expression of genes in treatment A and in treatment B. For example, many genes in treatment A (to the point of significant enrichment) may be regulated by transcription factors (TFs) X and Y while genes seen after treatment B are preferentially regulated by TFs D and Z.
In other words, try to group your genes into regulatory sets and ask if there is significant enrichment for that assignment for treatment A vs B and for each of those vs the entire genome.
Biological response: You can also move into the biological response. Do the differentially expressed genes elicit a biological response that makes sense with the cell biology and biochemistry also noted for these two treatments? This could be teased out with GO (gene ontology) analysis with a tool like DAVID or g:GOSt.
To clarify - let's say I have sets of microarray data from the treatment of mammalian cells with two different drugs (normalized to untreated controls). From some simple analyses, I suspect that the transcriptional response to these drugs is very different. For instance, the differentially-expressed genes in each treatment are non-overlapping. What else could I look at to demonstrate that these drugs are causing a very different transcriptional response in mammalian cells?
(In response to above):
MEFs + Drug A x 3 replicates;
MEFs untreated x 3 replicates
MEFs + Drug B x 3 replicates;
MEFs untreated x 3 replicates
what happens quantitatively to the extremes of your expression profile? can you show that some transcript which is strongly expressed under one of the drugs is only weakly, or not at all, expressed under the other treatment? I would go for that as a start.
Not clear what you mean by "genes differentially expressed in one sample aren't differentially expressed in the other." Differential expression means significantly different expression under one condition as compared with another (e.g. cancer versus normal, drug-treated versus control) - i.e. we compare groups of samples. So what are you comparing to what?
I agree with the comment above and also think it would be easier to help you with more details. What are you samples, conditions,... Can you detail a bit more please?
Need more details, for sure. What type of data? Replicates? What is the experimental design?