Supplement to: Predicting phenotypes from microarrays using amplified, initially marginal, eigenvector regression

For the DLBCL data, we not only list the selected genes, but also attempt to find any discussion of those genes in existing literature. Our final estimated model uses 49 gene features, which correspond to 26 genes. To examine the relevance of each selected gene for DLBCL, we adopt two approaches. The first endeavors to find literature examining the biological connection of the identified gene to any type of lymphoma. The second lists any reference in the (rather lengthy) methodological literature in statistics, computer science, and bioinformatics that uses statistical or machine learning methods to examine the DLBCL dataset. We display our findings for all 26 genes in Table 1. To summarize, 16 out of the 26 genes have been related to lymphoma in the biological literature, and 19 of them have already been identified via statistical techniques developed for the DLBCL dataset. While many of the 26 genes have been previously connected to lymphoma in general and DLBCL in particular, AIMER does identify 4 genes with symbols ALDH2, CELF2, COL16A1, and DHRS9 that have not been previously identified in the biological or methodological literature. We note that, while we have made every effort to locate each gene, given the large and evolving literature on this topic, those we have been unable to locate may have none-the-less been previously studied.