A systematic evaluation of normalization methods in quantitative label-free proteomics.
Välikangas T, Suomi T, Elo LL. · 2018
Read A systematic evaluation of normalization methods in quantitative label-free proteomics. by Välikangas T, Suomi T, Elo LL. (2018) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc5862339.txt.
Introduction
Välikangas and co-authors evaluate normalization methods for label-free proteomics. Their study appeared in Briefings in Bioinformatics. Proteomics data from mass spectrometry carry built-in bias. Sample handling and instruments both add unwanted variation. Normalization aims to make samples fairly comparable. The team tested several popular methods on four datasets. Three datasets used spike-in samples, and one used mouse tissue.
The authors report that variance stabilization cut replicate variation most. It also did well in tests for changed protein levels. Linear and local regression methods also performed well throughout. The team checked global against segmented normalization for analysis. They close with advice on what makes a method suitable. Careful choice protects all steps that follow.
Catalog record
| CREATOR | Välikangas T, Suomi T, Elo LL. |
|---|---|
| YEAR | 2018 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 47 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A systematic evaluation of normalization methods in quantitative label-free proteomics By Tommi Välikangas, Tomi Suomi, Laura L Elo Published in: Briefings in Bioinformatics DOI: 10.1093/bib/bbw095 ABSTRACT Abstract To date, mass spectrometry (MS) data remain inherently biased as a result of reasons ranging from sample handling to differences caused by the instrumentation. Normalization is the process that aims to account for the bias and make samples more comparable. The selection of a proper normalization method is a pivotal task for the reliability of the downstream analysis and results. Many normalization methods commonly used in proteomics have been adapted from the DNA microarray techniques. Previous studies comparing normalization methods in proteomics have focused mainly on intragroup variation. In this study, several popular and widely used normalization methods representing different strategies in normalization are evaluated using three spike-in and one experimental mouse label-free proteomic data sets. The normalization methods are evaluated in terms of their ability to reduce variation between technical replicates, their effect on differential expression analysis and their effect on the estimation of logarithmic fold changes. Additionally, we examined whether normalizing the whole data globally or in segments for the differential expression analysis has an effect on the performance of the normalization methods. We found that variance stabilization normalization (Vsn) reduced variation the most between technical replicates in all examined data sets. Vsn also performed consistently well in the differential expression analysis. Linear regression normalization and local regression normalization performed also systematically well. Finally, we discuss the choice of a normalization method and some qualities of a suitable normalization method in the light of the results of our evaluation. Keywords: proteomics, normalization, label free, bias, differential expression, logarithmic fold change, quantitation, intragroup variation, reproducibility, mass spectrometry INTRODUCTION The development of mass spectrometry (MS)-based proteomics has been rapid. Modern proteomics aims not only to identify the proteins but also to quantify them as accurately as possible [1]. Current MS-based proteomics workflows are able to detect thousands of proteins, their modifications and localizations in a single run [2]. Despite all the developments of MS technologies, the data from the MS analysis are still susceptible to systematic biases [3]. This bias has been defined as variation caused by nonbiological sources, which is introduced by small variations in the experimental conditions in the course of carrying out the MS analysis [4]. These variations include, for example, differences in sample preparation and handling, device calibration or changes in temperature, but the exact reason of the bias is usually unknown and cannot thus be solely accounted for by adjusting the experimental settings [3, 4]. The observed bias can be independent or dependent on the measured protein abundances [4]. The process that aims to take the bias into account is called normalization. Normalization aims to make the samples of the data more comparable and the following downstream analysis reliable [3]. Many of the normalization methods used for proteomics data have their roots in the DNA microarray technology [4], where several evaluations and reviews have already elucidated their performance [5–8]. For instance, Bolstad et al. [5] compared five normalization methods with DNA microarray data and concluded that most of them performed rather similarly and reduced nonbiological variability across arrays when compared with the unnormalized data. Choe et al. [6] also found no significant differences between the four normalization methods they examined with an RNA spike-in experiment at the probe level. In previous comparisons in proteomics, Callister et al. [9] used three different liquid chromatography–MS (LC-MS) data sets to evaluate four different normalization methods on peptide level and found a linear regression normalization best suited for their data sets. Kultima et al. [10] compared 10 different normalization methods with three different peptidomics data sets and noticed that the order of the LC-MS experiments affected the bias in the data; they suggested that their novel RegrRun normalization, which combines linear regression normalization with analysis order normalization, was the best overall method in reducing unwanted intragroup and intrasample variation. Different tools for helping in the selection of a normalization method have also been proposed. Webb-Robertson et al. [11] stated that a single method cannot account for the bias in different data sets; rather it is crucial for reliable downstream analysis to select the appropriate normalization method for each data set. They introduced a tool called SPANS, which combines eight methods for peptide selection to be used in normalization with five normalization methods [11]. Chawade et al. [3] also introduced a tool for choosing a proper normalization method called Normalyzer. Their tool includes several popular normalization methods such as linear regression, local regression, total intensity, average intensity, median intensity, variance stabilization normalization (Vsn) and quantile normalization, together with several frequently used evaluation measures used to assess the performance of a normalization method such as the pooled coefficient of variation (PCV), the pooled median absolute deviation (PMAD) and the pooled estimate of variance (PEV) [3]. So far, comparisons of normalization methods in proteomics have typically focused on their ability to decrease intragroup variation between technical and/or biological replicates of the test data. Measures for the intragroup variation such as PEV [3, 9, 10], PCV [3], PMAD [3], the median coefficient of variation (CV) [9] and the median SD [10] have been used to rank the normalization methods compared. While reducing intragroup variation is certainly a central goal of normalization, a more thorough comparison of the normalization methods and their performance in proteomics is still lacking. Although interesting questions such as differences in the correct detection of truly differentially expressed proteins in the data normalized by different normalization methods have been investigated before [3, 12, 13], a thorough systematic analysis using multiple data sets and two-group comparisons has not been available in proteomics. Also, the effect of the normalization method on the estimation of the logarithmic fold change (logFC) or the effect of how the normalization is performed when comparing only two sample groups from a larger data set has not been systematically investigated before. To address this need, we conducted an extensive comparison of 11 popular normalization methods or their variants. Other normalization approaches not covered in this study exist, such as the MaxLFQ integrated into the MaxQuant software [13] and the normalization integrated into the DeMix-Q software [14]. These normalizations, however, are integral parts of proteomics software workflows as opposed to the stand-alone normalization methods examined in this comparison, with the exception of Progenesis normalization. All the normalization methods examined are commonly used methods in proteomics and have different approaches and assumptions regarding the bias occurring in the data. Three spike-in label-free proteomics data sets were used for benchmarking the normalization methods. The spike-in data sets are suitable for this kind of method testing, as the differences between sample groups are known, and methods can be evaluated in their ability to find the true differences and to level out other biologically nonexisting differences. Additionally, a data set from a mouse study was also used to compare the performance of the normalization methods in a non-spike-in data set, representing a typical real research setting. Offline fractionation, which adds another layer of complexity to normalization, was not used in any of the tested data sets. In such cases, the total peptide ion signals of each