Calculating Sample Size Estimates for RNA Sequencing Data

[RockChalkJayhawk]

I know this was picked up by the Newsbot, but I would like to mention that we created an R package on bioconductorthat can calculate sample size, power, effect sizes, etc. The paper is located here and the link to the Bioconductor package is here

Feel free to let me know if you have any issues.

[dpryan]

Out of curiosity, how does RNAseqPower compare to Scotty when the same pilot data is used for both? Regardless, that's a nice tool to have!

[RockChalkJayhawk]

I hadn't actually seen it before, but it is a very good looking web app! The power estimates should be similar, but won't be identical for 2 reasons:

1. Scotty assumes that variance follows a lognormal distribution. I think this is a valid and logical assumption. RNASeqPower uses the measured variance from the data instead.

2. Scotty pseudo-randomly selects 200 values from a 2 dimensional matrix (variance and depth) then uses that mean as the power of the dataset. The way we use RNASeqPower is to say that we want to know the minimum requirements to reach the desired power. We need to know two things, how many genes do I want to detect and how much variance to I want to allow. If you only want to detect the top 90% of genes, then you take the 10th percentile of gene counts for your pilot (all other genes will have more than this count). You do the same for variance except you select the 90th percentile of variance. Using the minimum read count and the maximum allowable variance, that is the power you have in your experiment. Granted most genes will have higher power since they will have lower variance and higher counts. So in the end, it comes down to how you define what power actually means.

RNASeqPower can also give you the power for each gene in a dataset.

[dpryan]

Cool, thanks!