Classification of heterogeneous biological samples based on gene signature lists using a mixture of two normal distributions

Source: R/externalFunctions.R

This function classifies heterogeneous biological samples based on gene signature lists, effectively isolating signature-positive samples through an automated statistical framework.

The workflow begins by calculating a Gene Set Variation Analysis (GSVA) score for each sample to quantify the relative pathway activity. A subsequent permutation step extracts sample-specific standard deviations to capture the variance. By combining these metrics, a unique normal distribution is established for each sample. Through an up-scaling step, values are randomly drawn from these individual distributions and pooled to construct a comprehensive mixture of normal distributions that represent the entire sample population. A hypothesis-testing approach in which the null hypothesis posits that a sample belongs to this alternative distribution with the lower mean is applied. Samples that successfully reject the null hypothesis are assigned to the signature classification.

runSubtypingBimodal(
  geneLists,
  expectedCountsMatrix,
  permRatio = 0.75,
  permNbr = 20,
  upscaleNbr = 10
)

Arguments

geneLists

a list containing signature gene lists. At least two signature gene lists must be present.

expectedCountsMatrix

a normalized gene expression data matrix with rows corresponding to genes and columns to samples.

permRatio

a numeric between 0 and 1 representing the number of samples that are retained for the permutation step. The rounded value of the number of patients multiplied by this numeric should be inferior to the total number of patients so that the permutation step is not done on the entire cohort. Default: 0.75.

permNbr

a integer, 5 or higher, representing the number of permutation sampling done. In addition, the number of permutations must be equal or inferior to the total number of unique permutations possible with the dataset considering the parameters selected by the user. Default: 20.

upscaleNbr

a integer, 2 or higher, representing the number of values taken for the normal distribution for each sample to run the up-scaling step. Default: 10.

Value

a list of class splitTypeResults containing the signature-specific classification for each sample as well as all the results generated at each step of the statistical workflow.

Author

Astrid Deschênes

Examples


## Loading signatures
data("signaturesDemo")

## Load demo normalized expected counts for 30 patients
data("expNormalCountsDemo")

## Fix seed for reproducibility
set.seed(121)

## Run classification on the 30 patients using 20 permutations on 75% of 
## the dataset, and 10 points per patient for the up-scaling step
results <- runSubtypingBimodal(geneLists=signaturesDemo, 
    expectedCountsMatrix=expNormalCountsDemo, 
    permRatio=0.75, permNbr=20, upscaleNbr=10)
#> number of iterations= 55 
#> number of iterations= 72