Skip to contents

normScore: a package for assessing normalization in proteomics data

normScore is an R package for evaluating and ranking normalization methods in proteomics data based on a composite score that integrates multiple performance metrics.

Check the following repository to a deepen view into the normScore structure: normScore-development.

Click the following link to access the pkgdown web page of the package.

Do not you like coding? Do not worry, you can use the normScore shiny app.

Background

Preprocessing is a crutial step in proteomics data analysis, including dealing with missing data, imputation and normalization, among others. The options and parameters selected during this initial process will impact the following results. Thus, it is necessary to make the best decisions based on solid evidence.

Regarding normalization, Chawade et al. 2014 had proposed a combination of graphic and metric assessments to decide which normalization method should be applied to our data, that were implemented in a very useful and easy-to-used package with a corresponding shiny app which makes the tool more accesible to all users, specially the ones without coding skills.

To use normalyzerDE, it is necessary to know how to interpert the different graphics and the meaning of the different metric, a fact that can be difficult to reach for most of the users that will only used proteomics in a specific part of their project. Moreover, the final decision is based on the qualitative criteria of the user, and the own assessment can become a difficult task when the performance of the different normalization methods varies a lot across the multiple criteria used to assess.

Consequently, we have selected some of the criteria for normalization assessment proposed at NormalyzerDE. For each criteria, we built a metric to rank each normalization based on their performance for that specific criteria. Then, all metrics were standardized and summed to obtained a composite, global score that ranks all the normalizations based on their performance across the different metrics.

We strongly recommend using this score in comination with the graphical representations provided by NormalyzerDE, and using it as a guide that helps in the selection of the best normalization method.

Installing normScore

normScore package can be install through GitHub directly, both the develop or the stable version:

# Develop
devtools::install_github("juliagcurras/normScore")

# Stable

Quick start

Three imputs are required to use normScore main function:

  1. Your raw proteomic dataset (with proteins in rows and sample names at colnames), without logarithmic transformation.

  2. A list with this dataset after applying different normalization methods (one element for each normalization)

  3. The desing matrix, that is, a data.frame with two columnas with the correspondence between sample names (first column) and group names (second column).

With simulateData function, some example data can be generated to test the main function as in the following example:

# Load library
library(normScore)

# Simulate example data
simData <- normScore::simulateData(
    nProteins = 100,
    seed = 9396
) # List with 3 elements: raw data matrix | Log-transformed matrix | Metadata

# Create a list with all normalizations
listData <- list(
    Log = simData$logData,
    Method1 = simData$logData + 0.1,
    Method2 = simData$logData + 0.5
)

Once you gathered all input data, normScore is ready to be executed!

# Compare normalization methods with normScore
normScore(
    normalizedDataList = listData,
    groupData = simData$metadata,
    rawData = simData$rawData,
    refGroup = NULL, # if non-stated, a random group will be picked as reference
    altGroup = NULL # if non-stated, a random group will be picked as alternative
)