Data with only 1 pair for DEG analysis

Question

[bharata1803]

Hello,

So, currently, I have 1 pair of normal-cancer data from RNA-seq. I already generated gene level read counts with Salmon. What I want to do is to check for a certain gene whether it is up or down regulated. So, is there any problem with just 1 pair of sample? What tools will be suitable for 1 pair sample like this (DESeq2, limma, edgeR,etc)?

Thank you for your answers and suggestions.

Answer 1

[Chirag Parsania]

Hi Bharata,

It doesn't matter how many pairs/samples you have in differential analysis. What matters is whether you have biological replicates or not for each of your sample. If you have biological replicates then check the correlation of those replicates. If they are correlating well (corr coeff. >= 0.9) great. You can go for diff analysis. I have good exp with DESeq2. It works well for me. You can use other package as well (limma edgeR)and then make decision on the basis of what biology you are getting. whether you are getting relevant genes with significant p val or not in diff exp list.

Cheers

Chirag

[bharata1803]

Thank you. So, for this data, there is no replicates. I only have 1 fastq file from normal cell and 1 fastq file from cancer cell. That's why I am currently don't know what to do.

[Chirag Parsania]

It's little difficult to convince people on your findings without replicates. Anyway you can run DESeq2 without replicates as well. But in that case there is no meaning of your p-value and statistics. To overcome this you can filter highly up and down genes, confirm those in genome browser (for eg IGV ) and perform downstream analysis with those confidant diff. exp. genes.

Chirag

[bharata1803]

Thank you for your suggestion. I will try that. Also, I want to ask about replicates. So, if there are 3 cancer samples but from different patient/person, can those samples can be consider as replicates or it should come from same patient/person? Until now, the RNA-seq published via NCBI mostly compare several normal-cancer pair but from different patient/person. I know this replicates problem is very important for statistical significance so I want to understand more about this.

[Chirag Parsania]

The answer is yes and no. Yes because if those cancers are same you can consider them as biological replicates. Again I would suggest you to calculate correlation between those patients samples. If samples don't correlate though cancers are same (in different patients) you will be in trouble. You can not consider them as biological replicates. The reason for not correlating possibly technical or real biological phenomena. We don't know. So far what I have observed is people do sequencing of same sample in duplicates (one patient) to check the technical variability vs biological phenomena. That will give you more confidant but definitely more expensive.

Chirag