Label Binary Phenotype Cells Based on Prediction Scores with Minimum Threshold

Classifies cells into binary phenotype groups ("Positive" vs "Other") based on prediction score differences between two phenotypic conditions. This enhanced version includes a minimum threshold constraint to ensure biological relevance and provides detailed reporting on classification outcomes.

Usage

LabelBinaryCells(
  pred_dt,
  pheno_colnames,
  select_fraction,
  test_method,
  min_threshold = 0.7,
  verbose = TRUE
)

Arguments

pred_dt

A data.table containing prediction scores for two phenotypic conditions. Must contain columns specified in pheno_colnames.

pheno_colnames

Character vector of length 2 specifying the column names for the two phenotypic conditions to compare. The second element is used as the reference group if not found with regex matching.

select_fraction

Numeric value between 0 and 1 specifying the fraction of cells to classify as "Positive". Default selection depends on the distribution characteristics.

test_method

Character string specifying the statistical test to use for normality assessment. One of: "jarque-bera", "d'agostino", "kolmogorov-smirnov".

min_threshold

Numeric value specifying the minimum score difference required for a cell to be considered "Positive". This ensures biological relevance by filtering out weak associations. Default: 0.7.

verbose

Logical, whether to print messages.

Value

The input pred_dt with three additional columns:

• diff - Numeric vector of score differences between the two conditions

• label - Character vector with cell classifications: "Positive" or "Other"

The function also provides detailed console output about the classification process and results.

Details

This function implements a sophisticated approach for binary cell classification that adapts to the underlying distribution of prediction score differences while enforcing a minimum threshold for biological significance:

Classification Strategies with Minimum Threshold:

• Non-normal distributions (p-value < 0.05): Uses quantile-based selection where the top select_fraction of cells by score difference are classified as "Positive", but only if they exceed min_threshold

• Normal distributions (p-value >= 0.05): Uses normal distribution quantiles to determine the classification threshold, adjusted upward if necessary to meet the minimum threshold requirement

Supported Normality Tests:

• Jarque-Bera: Tests for skewness and kurtosis deviations from normality

• D'Agostino: Extended normality test focusing on skewness

• Kolmogorov-Smirnov: Non-parametric test comparing empirical distribution to normal distribution

Note

The minimum threshold parameter (min_threshold) helps prevent over- interpretation of weak phenotypic associations and ensures that classified cells show substantial differences between conditions. The function provides comprehensive feedback about threshold adjustments and final classification statistics.

See also

jb.test.modified(), dagostino.test(), ks.test() for the underlying normality tests used in the classification process.

Other DEGAS: DoDEGAS(), LabelContinuousCells(), LabelSurvivalCells(), Vec2sparse(), predClassBag.optimized(), readOutputFiles.optimized(), runCCMTL.optimized(), runCCMTLBag.optimized(), writeInputFiles.optimized()

Examples

if (FALSE) { # \dontrun{
# Create example prediction data
pred_data <- data.table(
  condition_A = runif(1000),
  condition_B = runif(1000)
)

# Classify cells using D'Agostino test with minimum threshold
result <- LabelBinaryCells(
  pred_dt = pred_data,
  pheno_colnames = c("condition_A", "condition_B"),
  select_fraction = 0.1,
  test_method = "d'agostino",
  min_threshold = 0.7
)

# Check classification results
table(result$label)

# View the actual fraction of positive cells
prop.table(table(result$label))
} # }