Gene category testing problems involve testing hundreds of null hypotheses that correspond to nodes in a directed acyclic graph. The logical relationships among the nodes in the graph imply that only some configurations of true and false null hypotheses are possible and that a test for a given node should depend on data from neighboring nodes. We developed a method based on a hidden Markov model that takes the whole graph into account and provides coherent decisions in this structured multiple hypothesis testing problem. The method is illustrated by testing Gene Ontology terms for evidence of differential expression.

very interesting paper on new methodology and supporting theory. Treelets seem to tackle two important problems of modern data analysis at once. For datasets with many variables, treelets give powerful predictions even if variables are highly correlated and redundant. Maybe more importantly, interpretation of the results is intuitive. Useful insights about relevant groups of variables can be gained. Our comments and questions include: (i) Could the success of treelets be replicated by a combination of hierarchical clustering and PCA? (ii) When choosing a suitable basis, treelets seem to be largely an unsupervised method. Could the results be even more interpretable and powerful if treelets would take into account some supervised response variable? (iii) Interpretability of the result hinges on the sparsity of the final basis. Do we expect that the selected groups of variables will always be sufficiently small to be amenable for interpretation? 1. Treelets or hierarchical clustering combined with PCA. A main part of

Background: The development of expression-based gene signatures for predicting prognosis or class membership is a popular and challenging task. Besides their stringent validation, signatures need a functional interpretation and must be placed in a biological context. Popular tools such as Gene Set Enrichment have drawbacks because they are restricted to annotated genes and are unable to capture the information hidden in the signature’s non-annotated genes. Methodology: We propose concepts to relate a signature with functional gene sets like pathways or Gene Ontology categories. The connection between single signature genes and a specific pathway is explored by hierarchical variable selection and gene association networks. The risk score derived from an individual patient’s signature is related to expression patterns of pathways and Gene Ontology categories. Global tests are useful for these tasks, and they adjust for other factors. GlobalAncova is used to explore the effect on gene expression in specific functional groups from the interaction of the score and selected mutations in the patient’s genome. Results: We apply the proposed methods to an expression data set and a corresponding gene signature for predicting survival in Acute Myeloid Leukemia (AML). The example demonstrates strong relations between the signature and cancer-related pathways. The signature-based risk score was found to be associated with development-related biological processes. Conclusions: Many authors interpret the functional aspects of a gene signature by linking signature genes to pathways or