The paper provides a solution to a widely acknowledged and urgent statistical problem, regarding a most useful research technique in genetics, the use of which is spreading fast.

“In recent years new extremely large problems emerged in different branches of science, and we realized that the FDR approach is the only practical way to address them”

Another good reason for its fast popularity in citing is the relatively long time that has passed since we first talked publicly about our solution and when the paper appeared in print—the delay was our own fault.

The paper describes a new methodology for the statistical analysis of gene expression data using microarray technology.

The False Discovery Rate (FDR) has been offered as an alternative to the traditional approaches to multiple hypotheses testing. In recent years, new extremely large problems emerged in different branches of science, and we realized that the FDR approach is the only practical way to address them. In particular we argued that the FDR approach is very appropriate for the analysis of microarray data. One of us became deeply involved in the practical analysis of microarray datasets. She realized the need for further research into the appropriateness and implementation of the FDR methodology in the particularities of microarray analysis, as well as the importance of reporting the results to the community of potential users.

It may happen that a gene will be discovered as differentially expressed in two biological situations even though the difference is merely due to natural variability in process and measurement. Since microarray data includes genetic expression on thousands of genes, often in numerous biological conditions, it becomes too likely that a gene will be falsely discovered as differentially expressed. Traditional statistical methods to cope with this problem try to keep the probability of making even a single false discovery very low—at the expense of requiring a very high threshold for calling a gene differentially expressed.

The FDR approach instead controls the expected proportion of false discoveries among the discovered genes. The procedures offered in the paper do so under the special dependency structure typical to microarray analysis. They result in less stringent thresholds, offering researchers a pool of discovered genes on which to continue research, without wasting too large a proportion of their efforts on non-relevant genes.

Anat Reiner, Daniel Yekutieli, and Yoav Benjamini
Department of Statistics and Operations Research
The Sackler Faculty of Exact Sciences
Tel-Aviv University
Tel-Aviv, Israel