A Statistical\nApproach for Identifying the Best Combination\nof Normalization and Imputation Methods for Label-Free Proteomics\nExpression Data

Label-free proteomics expression data sets often exhibit\ndata heterogeneity\nand missing values, necessitating the development of effective normalization\nand imputation methods. The selection of appropriate normalization\nand imputation methods is inherently data-specific, and choosing the\noptimal approach from the available options is critical for ensuring\nrobust downstream analysis. This study aimed to identify the most\nsuitable combination of these methods for quality control and accurate\nidentification of differentially expressed proteins. In this study,\nwe developed nine combinations by integrating three normalization\nmethods, locally weighted linear regression (LOESS), variance stabilization\nnormalization (VSN), and robust linear regression (RLR) with three\nimputation methods: k-nearest neighbors (k-NN), local least-squares\n(LLS), and singular value decomposition (SVD). We utilized statistical\nmeasures, including the pooled coefficient of variation (PCV), pooled\nestimate of variance (PEV), and pooled median absolute deviation (PMAD),\nto assess intragroup and intergroup variation. The combinations yielding\nthe lowest values corresponding to each statistical measure were chosen\nas the data set’s suitable normalization and imputation methods.\nThe performance of this approach was tested using two spiked-in standard\nlabel-free proteomics benchmark data sets. The identified combinations\nreturned a low NRMSE and showed better performance in identifying\nspiked-in proteins. The developed approach can be accessed through\nthe R package named ’lfproQC’ and a user-friendly Shiny\nweb application (https://dabiniasri.shinyapps.io/lfproQC and http://omics.icar.gov.in/lfproQC), making it a valuable resource for researchers looking to apply\nthis method to their data sets.