Are there any data augmentation techniques for gene expression data that can help overcome the challenges associated with developing a classifier using small sample sizes?
I am working on a project involving biomarker identification followed by binary classification, the objective being to develop a disease screening tool. My data comes from clinical samples and is of reasonably high quality, with all samples having RIN scores around 9. Replicates of a subset of samples indicate that there is very little technical variation. Unfortunately, there is substantial biological variation, both in the patient metadata (which has loads of missing values) and expression data. The expression data consists of ~25K genes measured by microarray, ~50 training samples, and ~30 testing samples.
