Comparing gene expression between two samples

Question

[zizigolu]

Hi,

I have raw read count matrix of HTG EdgeSeq for two patients A and B; I have tumour of patient A cultured in one media and tumour of patient B cultured in two medias, so I have

Patient A tumour cultured on OGM media

Patient B tumour cultured on DMEM and OGM medias

How I can say if DMEM and OGM medias influence the tumour culturing or not as I don't have any replication?

I found deAna for Differential expression analysis between two sample groups but I am stopping by error

Any suggestion please?

[GenoMax]

I have tumour of each patient cultured in two media

Not two media. Same media but for obvious reason cultured independently.

Patient A tumour cultured on DMEM and OGM media

Patient B tumour cultured on DMEM and OGM media

How I can say if DMEM and OGM media influence the tumour culturing or not as I don't have any replication?

How can you say that since media is identical in both cases.

[i.sudbery]

I think that there are two medias - media 1: DMEM, media 2: OGM?

[zizigolu]

Sorry, DMEM and OGM are two different medias

Answer 1

[i.sudbery]

If your interest is in determining the effect of media on gene expression, then for your purposes the two different patients act as replicates. See my answer here: C: Replicates for RNA-seq from 1 cell line undergoing different treatments

[zizigolu]

Thank you, I just had to remove the DMEM media for patient A because of too low quality in this sample; So now I have patient A with OGM media and patient B with DMEM and OGM medias, still is there the way to find if two medias are different or not?

[i.sudbery]

Well the analysis will run, but you should be very sceptical of the results, as DESeq/EdgeR will be doing a pooled estimate of dispersion, which is far from ideal. To be honest, two replicates isn't good to start with, going down to effectively 1.5 is bad. In fact, its so far from ideal, I'd be temped to leave a poor quality sample in.

[zizigolu]

Sorry is there anyway to see if DMEM and OGM medias are effective on one patient?

[i.sudbery]

No. If you have only one patient, then you have no replicates, and you cannot do differential expression.

[zizigolu]

As I have 2000 genes, could I treated my genes as individuals and use t-test to see if these medias (as my groups) are significantly different?

[i.sudbery]

In affect using each gene as a replicate?

What are these 2000 genes? Do you have any reason to suspect that they should all move in one direction, by a similar amount?

If you have a set of genes up expect to go, for instance, up, then you might be able to conduct a sign test. If you normalise the complete genome using something like TMM, or upper-quartile normalisation, so that the average fold change across all genes is 0, then look at the signs of change for your 2000 genes. If there is no effect you would expect 50% of the genes to be up and 50% to be down. If all of the genes are up, then this points to a non-random effect (you'd need to show this wasn't the case for another set of genes, say all other genes, as a negative control). You could put a p-value on this using a test of proportions against a 0.5 expected proportion, or a fishers test against the contingency table of set membership ~ direction of change.

Of course in the absence of replicates (or even with replicates to be fair), you'd need to think very hard about what other possible sources of this effect were (e.g. differing GC bias, differing degredation profiles, differing polyA capture rates, etc).

[zizigolu]

Sorry, I have raw read counts of 2545 genes coming from HTG EdgeSeq Oncology Biomarker Panel. Among my samples I have a patient whos toumor has been cultured in two medias; DMEM and GOM; I want to know if these medias are different in terms of culturing toumor or not, so I have done t-test on 2545 genes with my two samples (DMEM and GOM). I obtained a highly significant difference but I am not sure if I am right or wrong

[GenoMax]

As a side note: you may want to look up existing publications for this panel using paper search tool on HTG site. Click on Assay Name box and then choose Oncology biomarker panel.

[i.sudbery]

I'm not entirely sure what that difference is telling you. If your pannel is properly normalised, then the average fold-difference across all genes should be 0. If you are using each gene as a seperate measurement and the two media as two different conditions in a t-test (so effectively you have a 2545 replicate two sample paired t-test), then you are saying that genes in one sample have higher counts than in the other. But all this says to me is that the data is not correctly normalised - probably that you've collected more reads in general from one of the samples.

You say that these two were "amongst" your samples. Does this mean that you have more samples that were only cultured in one media?

[zizigolu]

To be honest, I am not good in statistics and experimental design. You please imagine I have two medias and 2454 genes; I have normalized read counts with cpm following log2. My question is if these medias make any difference in terms of gene expression. So, I have used t-test to compare two medias. I have two patients but as one sample in one patient is too low in terms of quality, I have to compare these two medias by raw read counts of one patient. If you were me, how would you please deal with comparing these two medias?