Abstract:
|
A parameter of interest has an unknown prior density and the realizations of this parameter are unobservable. The observed data are assumed to follow a known distribution condition on that parameter. Under g-modeling method, we can estimate the prior density by assuming the prior density as an exponential family density. We present Kolmogorov-Smirnov test to investigate the difference in prior distributions with g-modeling estimation method. Under simulation, we demonstrated the ability of this test to detect the difference in distributions. With single-cell RNA sequencing data, we observed raw reads counts for each cell across genes and the mean proportion of reads from each gene is the unobserved parameter of interest. We applied g-modeling method to estimate the distribution of concentration rate. Then, we ran Kolmogorov-Smirnov test in single-cell RNA sequencing data with two groups of cells to detect differential expression. In addition, we fitted the MAST hurdle model with the same data set for comparison. Generally, our method is more powerful than MAST method since more significantly differentially expressed genes are detected.
|