Publication details
Assessment of Microarray Data Correlation Structure Influence on False Discovery Rate Procedures in R.
| Authors | |
|---|---|
| Year of publication | 2007 |
| Type | Article in Proceedings |
| Conference | 56th Session of the International Statistical Institute |
| MU Faculty or unit | |
| Citation | |
| Field | Applied statistics, operation research |
| Keywords | Microarray Data; False Discovery Rate; Correlation Structure |
| Description | For significance level 0.05 the EBT procedure is unambiguously outperforming the FDR procedures with respect to the number of correctly identified genes. However, due to the fact that the EBT approach does not take into account the type I error rate the number of correctly identified genes is accompanied with an increased number of false positives. We can conclude that EBT is performing well on data composed of complicated Normal mixtures without correlation structure. The relationship between the FDR and EBT procedures is not so obvious for the significance level of 0.10 (results not shown here). According to the number of correctly called genes both FDR and EBT procedures behave similarly when applied to correlated data models, otherwise EBT seems to be more efficient. Moreover we can see a general tendency of all procedures to perform poorly (i.e. selecting a far too low number of correct genes) in data sets of high complexity. |