Robust Estimation of Labeling Based High-Throughput Relative Protein Expressions Sample Clauses

Robust Estimation of Labeling Based High-Throughput Relative Protein Expressions. When the distribution of the relative protein expression levels contain one or more of the issues: non-Gaussian tails, regions of data sparsity, excess kurtosis, or asymme- try, standard distributional approximations such as the N (0, 1), or N (µ, σ2), do not perform adequately. We propose to develop methodologies that make maximum use of the available data; account for both within and between variations of each data layer; and robustly capture the empirical distribution of the relative protein expres- sion levels. To this end, we work within the framework of finite mixture models, and maximize the flexibility in data modeling by empirically inferring the full shape of the class-conditional probability distribution of each mixture component. Furthermore, we propose to investigate the utility of skew normal, skew Student’s t, and Generalized Hyperbolic (GH) distributions, as suitable approximations for the distribution of relative protein expression levels, when the said distribution suffers from one or more of the afore mentioned irregularities. To our knowledge, this is the first work to explore the utility of these distributions from an ‘omics’ data anal- ysis perspective. The fitting of these distributions is rigorously controlled for both goodness-of-fit; the number of fitted mixture components; and for stable estimates of mixture parameters. We discuss these methodologies using a step by step data analy- sis procedure; where for each protein, we start with estimating its relative expression ratio, and end with calculating its posterior probability of being non-differentially expressed.