How to deal with RNA-SEQ samples with different number of raw reads?

Question

[Whoknows]

Hi all,

I have a human RNA-SEQ project consists 4 samples; 2 condition each of them has 2 replicates. Three of four samples contain 20 millions paired-end reads however one of them consist 50 millions paired-end reads. Read length is same for all samples, 150bp.

I'd like to know, Is this reads difference will affect on my final result or not?

Is there any way to normalized data for DE step based on their library size?

Thanks.

Answer 1

[Devon Ryan]

You only tend to run into problems once you get ~10x differences in depth. All common RNAseq tools (DESeq2, edgeR, limma/voom, etc.) will already properly normalize your data.

[Whoknows]

thanks devon, so 30 millions extra reads is not a big deal for common analysis pipeline, right?

[Devon Ryan]

Correct, though it's not so much the difference between them as their ratio that ends up being a problem.

[rezashiralmohammad]

Hello,

I have faced the exact problem but greater than it has mentioned above. My project is included 2 conditions (tumor vs. normal) with 21 and 5 replicate respectively. 4 samples in tumor condition contain 250-300 million reads however other samples consist of 60-100 million reads on average. Since the results show significant differences between these samples it seems DESeq2 normalization does not help. Is there any suggestion for fixing this problem?

Thanks.

Reza

[Devon Ryan]

This should be its own question.

[rezashiralmohammad]

Thanks. I submitted it as a new question.