Statistical analysis of the phosphoproteome - applied to signalling pathways affected by dioxin treatment

摘要

A high number of missing values in proteomic datasets requires careful statistical analysis. We could show that the bayes t-test and the rank product test are less sensitive toward missing values. Both methods are suitable for proteomics data with a low number of biological replicates where the classical t-test is inadequate. Bayes t-test and rank product method are complementary. We have shown a better performance of these combined tests for a large-scale phosphoproteomics dataset, by comparing the results for increasing missing values as carried out with the Golub set.