Gaining Accuracy for Gene Expression Data by Parsimonious Models

Abstract

Gene expression data must be accurate in order to promote extensive, reliable, and repeatable results and to compare treatments with few false positives and false negatives. One way to gain accuracy is by advanced experimental techniques, and another is by good experimental design, including replication. But these may not be enough to achieve even one significant digit, as shown by an example using oat data. This article introduces an additional opportunity to increase accuracy that involves parsimonious models, which has not yet been implemented in the gene expression literature to the best of our knowledge. Basically, a parsimonious model gains accuracy by selectively recovering signal in its model while selectively relegating noise to a discarded residual. Typically, this accuracy gain is equivalent to averaging over several times as many replications, but its cost is trivial, merely some computation. Consequently, this neglected way to gain accuracy is quite cost effective. For gene expression research, accuracy gain by parsimonious models should be a standard component of best practices.