Annotation of Genes/Proteins involved in Spermatogenesis
Greedy AUC Stepwise (GAS) is a novel algorithm, which maximize the AUC for given model. For a K-sparse problem, we assume the numbers of non-zero features should be less than K. Some authors proposed statistic learning to solve such problem such as LASSO, LARS, Compressive Sensing and so on. GAS is different from such regularity methods. When the model is given, such as SVM and logistic regression, GAS will find the best features using a special way of greedy searching.
The probability was calculated by GAS algorithm, ranging from 0 to 1. The closer it is to 1, the more possibly it functions in spermatogenesis. Without enough prior information to optimize the critical point, we suggest to use 0.5 as the cut-off. At that point, both the FPR and FNR are relative low according to our simulation study.