Entering edit mode
The paper "Oldham, M.C., P. Langfelder and S. Horvath, Network methods for describing sample relationships in genomic datasets: application to Huntington's disease. BMC Syst Biol, 2012. 6: p. 63." Provided a method called SampleNetwork to identify sample outliers, but the URL doesn't work anymore, thus I can't find the code to do my analysis. Can anyone give me some help?
The not working URL: http://www.genetics.ucla.edu/labs/horvath/CoexpressionNetwork/SampleNetwork.
Another question concerned SampleNetwork. Should I apply SampleNetwork to unnormalized data or normalized data. As mentioned in the tutorial:
"To derive maximum utility from the function, we prefer to apply SampleNetwork to gene expression data that have been minimally processed in a consistent fashion, as described in Materials and Methods from the journal article."
It seemed that we'd better apply SampleNetwork followed by normalization.
But in the paper "Oldham, M.C., P. Langfelder and S. Horvath, Network methods for describing sample relationships in genomic datasets: application to Huntington's disease. BMC Syst Biol, 2012. 6: p. 63.", in the Microarray data pre-processing part, the data pre-processing steps:
"Expression values were generated in R using the “expresso” function of the “affy” package (http://www.bioconductor.org/) with “mas” settings and no normalization, followed by scaling of arrays to the same average intensity (200)."
It seems that the author first did a scaling steps and then using SampleNetwork to identify outliers.
Further, in the paper "Gandal, M.J., et al., Shared molecular neuropathology across major psychiatric disorders parallels polygenic overlap. 2018. 359(6376): p. 693 - 697.", they also first used RMA followed by SampleNetwork.
I'm wondering whether sample filtering should come before or after normalization, and why?
I personally run low-level preprocessing and normalization (RMA or MASS5 for Affy chips; for RNA-seq, I would normalize the data and perhaps do a log- or variance stabilizing transformation) first, then (the equivalent of) SampleNetworks. SampleNetworks does sample filtering for you; I don't think it does probe/feature filtering, so you may want to do that yourself. Both before and after SampleNetworks are valid, although they may give you slightly different results.
Peter