Abstract:
|
Pathway analysis has been widely applied to elicit useful pathway information for further understanding of disease etiology or examination of associated genes. Current methodology performed analysis on each pathway separately and then ranked the pathways based on p-values or combination of p-values. Such comparison of p-values or functions of p-values is indirect and may not be fair, because all pathways of interest are not included in one single model simultaneously and genes appearing in more than one pathway can cause correlation among them. In this study we consider all competing pathways at the same time, incorporate possible correlation among pathways, and propose to prioritize pathways based on Bayesian posterior probabilities. We apply this approach on simulations and next-generation sequencing data of a breast cancer study. Comparisons with current methods such as the hypergeometric model and Signaling Pathway Impact Analysis (SPIA) for pathway rankings are conducted as well. The results show that the proposed model can prioritize competing pathways. This prioritization can provide a reference list that laboratory needs to be focused on simply a small number of pathways.
|