Pathway Detection Based on Hierarchical LASSO Regression Model

Rapid and accurate identification of potentially interested pathways through the analysis of genome-wide expression profiles remains an important challenge in bioinformatics. Most existing methods are based on hypothesis testing, such as GSEA. These methods mainly focus on individual pathways and rank them based on their individual strengths. However, biological pathways often work together to function. Therefore, it is important to consider their correlations in detection of pathways that are most closely related to the phenotypes. Considering this problem in the framework of variable selection, we propose a hierarchical LASSO regression (HLR) model to detect differentially expressed gene pathways, which automatically takes into account the correlation structure among the genes via regression. This approach is able to both select important gene pathways and remove unimportant genes within selected pathways. Both simulation and real data analysis show promising results.