Development of statistical and computational approach for preprocessing and analysing high-throughput proteomics data with missing values

Abstract

With the introduction of high throughput technologies such as mass spectrometry, proteomics data can be reliably generated from samples that can be further analysed using various statistical approaches. Various steps are involved in the proteomics data analysis such as data pre-processing (cleaning, filtering, normalization, imputation), differential expression analysis, data visualization, etc. However, proteomics data suffer from the problem of data heterogeneity and missing values. The proteomics data needs to be appropriately pre-processed before performing any statistical analysis and further downstream analysis. Various methods exist for normalization and imputation of proteomics expression data, each having their own limitations. In this project, we have developed shiny apps for visualization of raw MS data and protein / peptide identification data. Further, we have developed an improved normalization approach for normalization of proteomics data and the performance of the proposed normalization approach was compared with the existing methods [RobNorm, quantile normalization, variance stabilizing normalization and Loess Normalization (cyclic)]. We have developed hybrid approaches of imputation to impute missing values in proteomics expression data. The performance of the developed approaches of imputation was found to be better or at par when compared with the existing methods [e.g., KNN, SVD, KNN-QRILC, etc.]. Furthermore, we have developed an interactive shiny app for detecting differential expressed proteins considering heterogeneity and missing values, and accommodating complex experimental design. The methodology/ algorithm/ shiny apps developed under this project would be very useful to researchers and students involved in bioinformatics and for those interested in development of new methodologies/ algorithms/ approaches for normalization, imputation, and differential expression analysis of proteomics expression data. Further, the approaches can be applied to the data obtained from similar experiments (e.g., Metabolomics data). This will help for better analysis of the proteomics data obtained from high-throughput biological experiments.