7. Discussion
This paper suggests a quantitative framework to evaluate the effects of variance stabilization methods on detection of differentially phosphorylated peptides. This framework is not limited to variance stabilization methods; any preprocessing or normalization method can be evaluated by considering its effect on peptide classification.
The proposed kinome data generator simulates kinome microarray data that consists of foreground and background intensity value pairs. However, the proposed methodology can be used in situations where only foreground values are available. To simulated such a data, one should assume availability of backgrounds and assign zero to all background intensities.
In this paper we used t-test as a tool for detection of differentially phosphorylated peptides. Other detection methods are possible. Hence, a modified version of performance evaluation procedure to compare various methods for detecting differentially phosphorylated peptides is a suggestion for further research.
The visual representation of the VSN transformed arrays reveal that the VSN does not maintain fold-change values. Study of the effect of VSN transformation on various seeded fold-change values is suggested for future research. In addition, for the normalization technique, devising a method to rapidly find a fold-change in transformed data that is equivalent to a given fold-change in untransformed data would also be useful.