How many lanes do I need?

[[email protected]]

I will be performing an RNA-seq experiment for differential gene expression with Illumina HiSeq 2500 (100 or 150bp pair-end). My experiment is designed as 6 control conditions and 6 treatments. So with 3 biological replicates each, it gives an amount of 36 samples. I know HiSeq 2500 yields around 190-240 million reads per lane.
Now, the problem is that I don't have much idea of how many reads per sample or per condition I would be needing. Thus, I also don't know how many lanes would be necessary for these 36 samples.
I work with filamentous fungi and the aim is to evaluate whole-transcriptome changes. Maybe it has to do with the size of the target genome (the genome is already sequenced), but in which relation exactly? Can someone please help me with this?

[GenoMax]

Generally for human samples (3 gb genome) one would want to target around 25-30 Million reads per sample. So use that as an approximation to calculate the number of reads you would need per sample.

Illumina makes a read calculator application available on the web here. Try that as well.

[[email protected]]

Yes, thanks! I have used this coverage calculator and it recommends 25 million reads per sample for a human genome. If my reference genome is for instance five times smaller, I wonder if reducing the number of reads 5 times would be a fair approach.

[GenoMax]

That would be fine.

Edit: SNPSarus makes a good point below. If the gene density of your organism is similar then an equivalent reduction would be fine. Otherwise scale accordingly.

[SNPsaurus]

I would caution against using the ratio of your sample's genome size to a human genome for calculating how many reads are needed for a gene expression study. The # of genes may not linearly change with genome size. For instance, stickleback and human both have 20,000 annotated genes (Ensembl) even though stickleback is 1/6th the genome size.

You might want to outsource your sequencing to a place with a NovaSeq that could do the entire 36 samples in a single S1 lane and should compare favorably on price to 3 lanes of a HiSeq 2500.

[[email protected]]

Thank you both, I really appreciate it!
I've been considering a comparison based on the number of genes. The fungi species I'm working on has around 8,400. So the 25 million reads per sample for minimum coverage of the human transcriptome (20,000 genes) would correspond to 10.5 million reads in my species.
Just for safety, I will keep my sequencing above 12 million so I don't have to worry.

[SNPsaurus]

As an additional complication, you might consider that a human sample is likely to be a mixture of cell types and the sequencing amount is designed to pick up expression of moderately-active genes in cells that may be a small fraction of the population. Your fungal samples might not be as diverse. At 8M reads per sample you have 1,000 reads per gene per sample so would still get 10 reads if the gene is 10-fold lower expression than average in a cell state that is 10% of the population. That is pretty sensitive already.

Countering that is that getting even reads across samples is not perfect and you might expect most samples to be in a 2-fold range with some outliers beyond that. Anyway, good luck on your experiments!

[[email protected]]

Well pointed, and actually I will be working with cells that are theoretically identical or almost identical. As my goal is to have a global picture of processes modulated by treatment conditions, I guess I just need to make sure to avoid an excessively low sequencing coverage.
Two lanes of HiSeq 2500 will be more than enough for that. One would not suffice. Unfortunately, I couldn't find any NovaSeq with a good price nearby, but I'll keep searching.