any tools to test if my RNAseq data is enough?

[mihuzx]

Hi,
I am new to the field of RNA sequencing.I have read a lot of message about how many reads is enough for RNAseq,but it seems that there is no exact answer.
Anyway, now I just wondering if I get the RNAseq data,such as 2Gb file, and How I can make sure it is enough for my experimernt.
maybe I can test that as follows:


but it is too difficult for me to draw this map. Does anyone know how to draw this picture,or have other tools to do this work!
very appreciate for any reply.

[dpryan]

It looks like you uploaded the wrong image.

Anyway, if you're curious how much more/less sequencing you could use, you could just make a rarefaction plot.

[mihuzx]

It looks like you uploaded the wrong image.

Anyway, if you're curious how much more/less sequencing you could use, you could just make a rarefaction plot.

thank you very much. I am sorry for the wrong picture, but it just is a rarefaction plot.

is there any tool to draw a rarefaction plot? I am not good at computer language ,any software or script is best!

Thank you very much!

[dpryan]

I don't think there's any push-button solution that would be very reliable, it'd be best to script something to best match your needs.

[mihuzx]

I don't think there's any push-button solution that would be very reliable, it'd be best to script something to best match your needs.

thank you for your reply, and I do need to learn more about computer languages ,such as R,perl or something...

[blancha]

If you have replicates and use Cuffdiff, you can use CummeRbund to plot the variation between replicates at different FPKM values, which can give an indication of the sequencing depth required.

CummeRbund is an R package, with built-in functions to generate plots that are useful to interpret your data.

[Wallysb01]

There are some general guidelines out there. Encode, IIRC, suggests at least 3 replicates per condition and 20M reads per replicate for gene level testing. Some other work though suggest that much lower sequencing depth and more replicates may be more useful (i.e., 6 replicates at 10M reads each). For isoform level tests, novel isoform detection or fusion gene detection, you may want much more depth (like maybe 60M read or more).

[mihuzx]

There are some general guidelines out there. Encode, IIRC, suggests at least 3 replicates per condition and 20M reads per replicate for gene level testing. Some other work though suggest that much lower sequencing depth and more replicates may be more useful (i.e., 6 replicates at 10M reads each). For isoform level tests, novel isoform detection or fusion gene detection, you may want much more depth (like maybe 60M read or more).

thank you, It is really very useful for me!

[mbblack]

Hi,
I am new to the field of RNA sequencing.I have read a lot of message about how many reads is enough for RNAseq,but it seems that there is no exact answer.
Anyway, now I just wondering if I get the RNAseq data,such as 2Gb file, and How I can make sure it is enough for my experimernt.
maybe I can test that as follows:


but it is too difficult for me to draw this map. Does anyone know how to draw this picture,or have other tools to do this work!
very appreciate for any reply.

An off the cuff answer might be to simple get as much read depth as you can possible afford. With RNA-seq and differential gene expression, more is quite literally always better. The issue is that beyond somewhere between 10-25M reads per sample (depending on your samples, and your choice of NGS technology), you get into a zone of very rapidly diminishing gains relative to cost and effort. You end up primarily piling up reads on already well characterized transcripts, adding very few new reads to low count transcripts, and detecting fewer and fewer "new" or previously undetected genes. So while literally "more is better", collecting more than 10M-20M mapped reads per sample will rapidly become prohibitive and for each incremental gain in total reads, your gain in truly novel informative data will get ever smaller (as all you mainly are doing is adding further to genes you alreay have very well characterized).

However beware, that your most challenging genes to characterize are those with relatively low expression and hence relatively few mapped reads. Those genes will have by far the largest variance, meaning it is difficult to collect enough data for them to have statistically significance. The fewer the reads, the harder it will be to detect significantly differentially expressed low expressors (it's never easy with those genes anyway, but more data can help).

Also do not neglect biological replication. 3 reps would be the bare minimum anyone should be using for DGE studies, but again, more is definitely better and will give more reliable statistical results.