Resampling Based Methods for Identifying Differen Sample Clauses

Resampling Based Methods for Identifying Differen tially Expressed Proteins Using XIC Area The observed protein profiles of two technical replicates of the same protein mixture can differ both with respect to the number of proteins identified and their expres- sion levels. The latent error mechanisms leading to these discrepancies are not well understood. However, it is believed that experimental and/or physicochemical prop- erties of the constituent peptides of the proteins themselves might account for much of this variability. This variability can be reduced to some degree through careful quality control at each stage of the experiment; by removing ‘unreliable’ data points; and by using a more stable method to quantify the peptide level relative expression ratios. In this context, we propose to use a bivariate mixture model based cluster- ing analysis to cluster the data into ‘reliable’ and ‘unreliable’ groups; and to use Xxxxxxxx-Xxxxx filtered ion current (XIC) profile area ratios to quantify the relative expression ratio of peptides. We propose to account for the latent error mechanisms by a novel method that relies on drawing weighted nested-bootstrap samples, and using a bootstrap partial maximum likelihood estimator (BPMLE) to estimate the overall relative expression ratio of each protein. We also develop a novel method that constructs a model - based BPMLE, through resampling the residuals from a robust Xxxxx regression of peptide level relative expression ratios. Finally, we assign a sig- nificance value to each estimated protein expression ratio using a nested-bootstrap p-value calculation, and then control the false discovery rate locally by modeling the p-value distribution as a two component Beta mixture model.