Abstract:
|
In genomics applications, we are interested to measure the differential expression level of genes between case and control groups. In order to determine that level, we estimate the value of parameter of interest(POI)which is a fold change between those two groups. There are several estimators of the POI, among them unshrunken and shrunken estimators. The empirical Bayes theory used to estimate the shrunken estimators by computing the posterior mean of POI, which is based on the unshrunken estimator and an estimator of local false discovery rate(LFDR).The LFDR is the posterior probability of a null hypothesis given a statistic test. Our aim is to study and compare the different estimators of POI by using the different estimators of LFDR like: a histogram-based estimator(HBE), a binomial-based estimator(BBE), a maximum-likelihood estimator(MLE), a corrected false discovery rate(CFDR), a reranking false discovery rate(RFDR),etc. These POI estimators will be applied on breast cancer and colon cancer data. Our simulation study show that, two shrunken estimators of POI which are BBE and RFDR may be preferred since their mean square error(MSE) performs better than the others estimators.
|