Rank Normalization Methods Using a Composite Scoring Framework

Computes a composite score to rank normalization methods based on multiple quality metrics derived from normalized data matrices.

The function evaluates each normalization method using six criteria: pooled coefficient of variation (PCV), within-group sample correlation, MA-plot trend deviation, mean-SD trend deviation, relative log expression (RLE) consistency, and total intensity consistency. These metrics are scaled, combined into a final score, and used to rank the normalization methods.

In addition, a correction factor derived from the coefficient of variation of raw sample intensities is applied to the "Log" normalization method. Optionally, bootstrap resampling is used to estimate mean composite scores and percentile-based confidence intervals.

Usage

normScore(
  normalizedDataList,
  groupData,
  rawData,
  refGroup = NULL,
  altGroup = NULL,
  returnDetails = TRUE,
  nBoot = 1000
)

Arguments

normalizedDataList: Named list of normalized numeric matrices or data frames. Each element represents one normalization method, with rows corresponding to features (e.g. proteins) and columns to samples.
groupData: data.frame or matrix describing the sample-group assignment. It must contain two columns, which are interpreted as sample names and group labels.
rawData: data.frame or matrix of raw numeric intensities, with rows representing features and columns representing samples.
refGroup: character or NULL. Reference group used for the MA-plot metric. If NULL, the last group in groupData is used.
altGroup: character or NULL. Alternative group used for the metric. If NULL, the first group in groupData is used.
returnDetails: logical. If TRUE, only the final ranking is MA-plot returned. If FALSE, detailed scores, bootstrap results, and a summaryplot are also returned. Default is FALSE.
nBoot: integer. Number of bootstrap resamples used to estimate mean normScore values and confidence intervals. Default is 1000.

Value

If returnDetails = TRUE, returns a list with:

finalRanking: Named numeric vector with the final normalization ranking.

If returnDetails = FALSE, returns a list with:

finalRanking: Named numeric vector with the final normalization ranking.
detailRanking: data.frame with scaled item-wise scores, total score, and corrected total score for each normalization method.
bootstrapScore: data.frame with bootstrap mean scores and 95\ percentile confidence intervals for each normalization method.
graphic: A ggplot2 object showing the bootstrap mean scores and confidence intervals.

Details

The function computes the following six item scores for each normalization method:

Item1: mean pooled coefficient of variation across groups (PVC).
Item2: within-group Spearman correlation summary, transformed so that lower values indicate better performance (Correlation).
Item3: shape-corrected area-based deviation from the expected MA trend (logFC = 0), based on MAplot.
Item4: area-based deviation of the mean-SD trend from horizontality, based on meanSD plot.
Item5: RLE-based mean absolute percentage error relative to 1.
Item6: quantile-based total intensity consistency metric.

Each item is scaled to the range 0 to 1 using min-max scaling and then summed into a total score. Lower scores indicate better normalization performance.

A correction factor based on the coefficient of variation of total raw sample intensities is applied only to the normalization method named "Log".

When returnDetails = FALSE, bootstrap resampling of the six item scores is performed.The results can be visualized using plotBootstrapNormScore.

Examples


# Simulate proteomic data
simData <- simulateData(nProteins = 1000)

# Normalyze ysing NormalyzerDE package
normalizedDataList <- list(
    Norm1 = simData$logData + 0.1,
    Norm2 = simData$logData + 1,
    Norm3 = simData$logData - 1,
    Norm4 = simData$logData * 1.1,
    Norm5 = simData$logData * 0.9,
    Norm6 = simData$logData * runif(ncol(simData$logData), 0.8, 1.2)
)

# Compute ranking
result <- normScore(
    normalizedDataList,
    groupData = simData$metadata,
    rawData = simData$rawData,
    refGroup = NULL,
    altGroup = NULL,
    returnDetails = TRUE,
    nBoot = 100
)
#> Reference group set to G2.
#> Alternative group set to G1.
#> Log2-transformed raw data were added to
#>             'normalizedDataList' as 'Log'.
result
#> $finalRanking
#>        Log      Norm5      Norm2      Norm1      Norm6      Norm3      Norm4 
#> 0.07079693 1.28431816 1.91416771 2.41162482 2.63643762 3.09313202 3.65587983 
#> 
#> $detailRanking
#>           Item1 Item2     Item3     Item4     Item5      Item6    Total
#> Log   0.4713758   0.1 0.5000000 0.8142192 0.4999485 0.08450405 2.470048
#> Norm5 0.4713758   0.1 0.0000000 0.6284383 0.0000000 0.08450405 1.284318
#> Norm2 0.0000000   0.1 0.5000000 0.8142192 0.4999485 0.00000000 1.914168
#> Norm1 0.4218242   0.1 0.5000000 0.8142192 0.4999485 0.07563292 2.411625
#> Norm6 0.4713758   0.0 0.8346731 0.0000000 0.3303888 1.00000000 2.636438
#> Norm3 1.0000000   0.1 0.5000000 0.8142192 0.4999485 0.17896431 3.093132
#> Norm4 0.4713758   0.1 1.0000000 1.0000000 1.0000000 0.08450405 3.655880
#>       TotalxItem0
#> Log    0.07079693
#> Norm5  1.28431816
#> Norm2  1.91416771
#> Norm1  2.41162482
#> Norm6  2.63643762
#> Norm3  3.09313202
#> Norm4  3.65587983
#> 
#> $bootstrapScore
#>   normalization meanNormScore      ll95       ul95
#> 1           Log     0.1393238 0.0545619  0.2674648
#> 2         Norm5     3.7364414 1.1994417  7.1184431
#> 3         Norm2     5.4469503 1.3080386 10.7193999
#> 4         Norm1     6.8978096 2.2940229 13.1223678
#> 5         Norm6     7.3844707 1.7794351 14.3342643
#> 6         Norm3     8.8855229 3.1746362 16.6470613
#> 7         Norm4    10.4002546 3.3071218 19.7737763
#>