RN_pmi {RNentropy}

Compute point mutual information matrix for the experimental conditions.

Description

Compute point mutual information for experimental conditions from the overxexpressed genes identified by RN_select.

Usage

RN_pmi(Results)

Arguments

Results

The output of RNentropy, RN_calc or RN_select. If RN_select has not already run on the results it will be invoked by RN_pmi using default arguments.

Value

The original input containing

gpv

-log10 of the Global p-values

lpv

-log10 of the Local p-values

c_like

a table similar to the one you obtain running the C++ implementation of RNentropy.

res

The results data.frame containing the original expression values together with the -log10 of Global and Local p-values.

design

The experimental design matrix.

selected

Transcripts/genes with a corrected Global p-value lower than gpv_t. Each condition N gets a condition_N column which values can be -1,0,1 or NA. 1 means that all the replicates of this condition seems to be consistently over-expressed w.r.t the overall expression of the transcript in all the conditions (that is, all the replicates of condition N have a positive Local p-value <= lpv_t). -1 means that all the replicates of this condition seems to be consistently under-expressed w.r.t the overall expression of the transcript in all the conditions (that is, all the replicates of condition N have a negative Local p-value and abs(Local p-values) <= lpv_t). 0 means that one or more replicates have an abs(Local p-value) > lpv_t. NA means that the Local p-values of the replicates are not consistent for this condition.

And two new matrices:

pmi

Point mutual information matrix of the conditions.

npmi

Normalized point mutual information matrix of the conditions.

Author(s)

Giulio Pavesi - Dep. of Biosciences, University of Milan

Federico Zambelli - Dep. of Biosciences, University of Milan

Examples

data("RN_Brain_Example_tpm", "RN_Brain_Example_design")
#compute statistics and p-values (considering only a subset of genes due to
#examples running time limit of CRAN)
Results <- RN_calc(RN_Brain_Example_tpm[1:10000,], RN_Brain_Example_design)
Results <- RN_select(Results)
Results <- RN_pmi(Results)

## The function is currently defined as
RN_pmi <- function(Results)
{
  if(is.null(Results$selected))
    Results <- RN_select(Results)
  
  Results$pmi <- matrix(nrow = ncol(Results$design), 
                        ncol = ncol(Results$design))
  
  colnames(Results$pmi) <- colnames(Results$design)
  rownames(Results$pmi) <- colnames(Results$design)
  
  Results$npmi <- Results$pmi
  
  colshift <- ncol(Results$selected) - ncol(Results$design) 
  
  for(x in 1:nrow(Results$pmi))
  {
    for(y in 1:nrow(Results$pmi))
    {
      if(x > y)
      {
        Results$pmi[x,y] <- Results$pmi[y,x]
        Results$npmi[x,y] <- Results$npmi[y,x]
        next
      } else
      {
        sum_x <- sum(Results$selected[,x+colshift] == 1, na.rm = TRUE)
        sum_y <- sum(Results$selected[,y+colshift] == 1, na.rm = TRUE)
        sum_xy <- sum(Results$selected[,x+colshift] == 1 & 
                        Results$selected[,y+colshift] == 1, na.rm = TRUE)
        freq_x <- sum_x / nrow(Results$selected)
        freq_y <- sum_y / nrow(Results$selected)
        freq_xy <- sum_xy / nrow(Results$selected)
        
        h_xy <- log2(1/freq_xy)
        
        Results$pmi[x,y] <- log2(freq_xy / (freq_x * freq_y))
        Results$npmi[x,y] <- Results$pmi[x,y] / h_xy
      }
    }
  }
  
  return (Results)
  }