macat (version 1.46.0)

scoring: Compute (regularized) t-scores for gene expression data

Description

This function computes for all genes in an expression matrix the (regularized) t-scores (statistics) with the given class labels and a number of permutations of these labels. Each gene is also assigned a p-value either empirically from the permutation scores or from a t-distribution.

Usage

scoring(data, labels, method = "SAM", pcompute = "tdist",  nperms = 1000, memory.limit = TRUE, verbose = TRUE)

Arguments

data

Expression matrix with rows = genes and columns = samples

labels

Vector or factor of class labels; Scoring works only with two classes!

method

Either "SAM" to compute regularized t-scores, or "t.test" to compute Student's t-statistic

pcompute

Method to compute p-values for each genes, either "empirical" to do permutations and compute p-values from them, or "tdist" to compute p-values based on respective t-distribution

nperms

Number of permutations of the labels to be investigated, if argument 'pcompute="empirical"'

memory.limit

Logical, if you have a really good computer (>2GB RAM), setting this FALSE will increase speed of computations

verbose

Logical, if progress should be reported to STDOUT

Value

A list, with four components:

observed

(Regularized) t-scores for all genes based on the given labels

pvalues

P-values for all genes, either from permutations or t-distribution

expected.lower

2.5%-quantile of permutation test-statistics, supposed to be a lower 'significance border' for the gene; or NULL if p-values were computed from t-distribution

expected.upper

97.5%-quantile of permutation test-statistics, supposed to be an upper 'significance border' for the gene; or NULL if p-values were computed from t-distribution

Details

If 'pcompute="empirical"', the statistic is computed based on the given class labels, afterwards for 'nperms' permutations of the labels. The p-value for each gene is then the proportion of permutation statistics that are higher or equal than the statistic from the real labels. For each gene the 2.5%- and the 97.5%-quantile of the permutation statistics are also returned as lower and upper 'significance threshold'.

If 'pcompute="tdist", the statistic is computed only based on the given class labels, and the p-value is computed from the t-distribution with (Number of samples - 2) degrees of freedom.

References

Regarding the regularized t-score please see the macat vignette.

See Also

evalScoring

Examples

  data(stjd)
  # compute gene-wise regularized t-statistics for
  #  T- vs. B-lymphocyte ALL:
  isT <- as.numeric(stjd$labels=="T")
  TvsB <- scoring(stjd$expr,isT,method="SAM",pcompute="none")
  summary(TvsB$observed)