Subject: Application of RUV to a small Nanostring dataset

Dear Forum,

I will soon be analyzing a small NanoString dataset of

1. 54 experimental genes, 2 housekeeping genes and the standard positive and negative controls. The experimental genes were chosen because they are likely to be differentially expressed, so that I should not use a normalization method which depends upon most of the genes not being differentially expressed.

2. 5 known conditions.

3. 4 biological replicates per condition.

4. To be run in a single batch.

Questions:

1. Are there possible advantages to normalizing RUV as distinct from NanoStringNorm on this small, one-batch dataset?

2. I have read the recent paper on RUV-III as applied to Nanostring:

https://www.ncbi.nlm.nih.gov/pubmed/31114909

Should RUV-III be applied to a differential expression experiment with known conditions, as distinct from one in which the factors of interest are unobserved?

After reading the paper, I am not clear on this point (my fault).

1. If RUV-III should be used do I need technical replicates for each condition, or for only 1 condition?

4, If RUV-III should not be used, but some kind of RUV should be used, which kind of RUV should be used, RUVSeq because Nanostring data is discrete or RUV for microarrays (RUV2 etc)?

Any guidance would be appreciated.I realize that the RUV package is in CRAN, but the RUVSeq package is in Bioconductor, and this forum seemed to me to be the most appropriate one.

Thanks and best wishes, Rich Richard Friedman, Columbia University

Since no one answered your question in 8 months I’ll give it a shot.

Since you do not have any technical replicates in your study then for sure you cannot use RUV-III. In the paper it was also not explicitly stated whether you needed technical replicates in each condition, only for it to work really well you needed technical replicates in each of your batches, if you have batches. But intuitively I believe this equally applies to each condition of interest.

What’s more applicable to NanoString data without technical replicates is to use RUVSeq, specifically RUVg. See thread https://support.bioconductor.org/p/123295/