A Gene Prognosis Profile using Partial Least Square (PLS)

In a typical clinical study based on a microarray gene expression experiment, there are often more genes than subject samples, and many genes are correlated. Partial least square regression, though not originally designed for classification purposes, can be used to build a classifier to predict outcomes based on the high dimensional correlated gene expression data. In this article, using a publicly available breast cancer study data, we show the process of using PROC PLS in SAS to construct a metastasis risk classifier from a training dataset. The classifier is further used to assign patients from an independent validation set into high- and low- risk groups based on their gene expressions. The result confirms that patients in the two predicted risk groups show significant difference in their survival outcome.