Finding Single and Multi-Gene Expression Patterns for Psoriasis Using Sub-Pattern Frequency Pruning

Biomarker identification, such as gene expression, is used in several areas of medical research, including aiding in disease prediction and treatment. However, most gene expression analysis focuses on differently expressed genes, ignoring patterns in which the co-expression of non-differently expressed genes are associated with disease risk. In this manuscript, we make three contributions. First, we present an alternative definition for differential expression which captures associations that are missed using mean- or median-based methods, such as fold change. Second, we introduce an algorithm for identifying all patterns of analytes associated with a given phenotype within a given threshold of optimal by extensively pruning the solution space. Third, our demonstration on psoriasis gene expression data yields 6320 highly significant gene expression patterns associated with this common disease that are comprised of 2334 unique genes worthy of further exploration. Interestingly, these genes include 1021 genes that are not differentially expressed when examined in isolation. Our approach is computationally efficient and our open-source software is freely available. This method holds potential for biomarker discovery for diverse phenotypes and is also applicable for identifying patterns hidden within non-biological real-valued data sets.