Simulate label-free proteomics-like two-group data

Simulates two-group quantitative omics data with controlled signal structure, including abundance-dependent variance, differential expression, sample correlation, global shifts, mean-SD dependence, and optional MNAR missingness.

Usage

simulateData(
  nProteins = 10000,
  nPerGroup = 20,
  muMean = 18,
  muSd = 1.2,
  muClip = c(15, 25),
  rhoWithin = 0.85,
  rhoBetween = 0.55,
  loadingSd = 0.25,
  sigmaHi = 0.05,
  sigmaLo = 0.4,
  gammaSigma = 2.5,
  propDE = 0.35,
  logFCSd = 1,
  logFCMean = 0,
  heteroLogFC = TRUE,
  fcHi = 0.55,
  fcLo = 2.5,
  gammaFC = 7,
  sampleShiftSd = 0,
  sampleShiftCap = 0.2,
  sampleSdStrength = 0,
  sampleSdRho = 0.8,
  sampleSdCap = 0.35,
  addMissing = TRUE,
  targetMissing = 0.001,
  kMnar = 1.2,
  missingBySampleSd = 0.05,
  seed = 9396
)

Arguments

nProteins

integer. Number of proteins or features to simulate. Default is 10000.

nPerGroup

integer. Number of samples per group. Default is 20.

muMean

numeric. Mean of the protein-wise abundance distribution on the log2 scale. Default is 18.

muSd

numeric. Standard deviation of the protein-wise abundance distribution on the log2 scale. Default is 1.2.

muClip

numeric vector of length 2. Lower and upper bounds used to truncate simulated protein means. Default is c(15, 25).

rhoWithin

numeric. Within-group correlation target for the latent sample factor. Default is 0.85.

rhoBetween

numeric. Between-group correlation target for the latent sample factor. Default is 0.55.

loadingSd

numeric. Standard deviation of protein-specific loadings for the correlated latent factor. Default is 0.25.

sigmaHi

numeric. Residual standard deviation at high abundance. Default is 0.05.

sigmaLo

numeric. Residual standard deviation at low abundance. Default is 0.4.

gammaSigma

numeric. Controls how strongly residual variance depends on abundance. Default is 2.5.

propDE

numeric. Proportion of differentially expressed proteins. Default is 0.35.

logFCSd

numeric. Standard deviation of differential expression log-fold changes. Default is 1.

logFCMean

numeric. Mean of differential expression log-fold changes. Default is 0.

heteroLogFC

logical. If TRUE, the variance of log-fold changes depends on abundance. Default is TRUE.

fcHi

numeric. Lower multiplier for abundance-dependent logFC heterogeneity. Default is 0.55.

fcLo

numeric. Upper multiplier for abundance-dependent logFC heterogeneity. Default is 2.5.

gammaFC

numeric. Controls how strongly logFC variability depends on abundance. Default is 7.

sampleShiftSd

numeric. Standard deviation of global sample shifts. Default is 0.

sampleShiftCap

numeric. Maximum absolute value allowed for global sample shifts. Default is 0.2.

sampleSdStrength

numeric. Strength of sample-specific mean-SD dependence. A value of 0 implies independence. Default is 0.

sampleSdRho

numeric. Correlation between sample mean rank and sample-specific SD effect. Must lie between 0 and 1. Default is 0.8.

sampleSdCap

numeric. Maximum absolute value allowed for the log-multiplier controlling sample-specific SD effects. Default is 0.35.

addMissing

logical. Should MNAR missing values be added? Default is TRUE.

targetMissing

numeric. Target overall proportion of missing Default is 0.001.

kMnar

numeric. Strength of abundance dependence in the MNAR missing values. value mechanism. Default is 1.2.

missingBySampleSd

numeric. Standard deviation of sample-specific missingness shifts. Default is 0.05.

seed

integer. Random seed used for reproducibility. Default is 9396.

Value

A list with the following elements:

logData

Numeric matrix of simulated log2-scale data.

rawData

Numeric matrix of simulated raw-scale data, obtained as 2^logData.

metadata

data.frame containing sample names and group labels.

Details

The simulated data include several structured components:

1. Protein-wise mean abundance on the log2 scale.

2. A latent correlation structure inducing stronger within-group than between-group sample correlation.

3. Abundance-dependent heteroscedastic residual noise to create an MA-like wedge pattern.

4. Symmetric differential expression between the two groups.

5. Optional global sample shifts affecting overall intensity.

6. Optional sample-level mean-SD dependence.

7. Optional MNAR missingness driven by abundance and sample-specific effects.

The output is intended for testing normalization methods and associated scoring procedures under controlled simulation settings.

Examples

simData <- simulateData(
    nProteins = 1000,
    nPerGroup = 5,
    propDE = 0.2,
    addMissing = TRUE,
    seed = 123
)

dim(simData$logData)
#> [1] 1000   10
head(simData$metadata)
#>   Samples Groups
#> 1    G1_1     G1
#> 2    G1_2     G1
#> 3    G1_3     G1
#> 4    G1_4     G1
#> 5    G1_5     G1
#> 6    G2_1     G2