R/getAdja.R

In lpNet: Linear Programming Model for Network Inference

#
# get adjacency matrix when solving LP problem with lpSolve
#
getAdja <-function(res, n, annot=NULL) {

  edges <- cbind(res$solution, names(res$objective)) 
  exist_edge <- edges[which(edges[ ,1] != 0), ]
  adja <- matrix(0, nrow=n, ncol=n)

  if (!is.null(dim(exist_edge))) {
		id_neg <- grep("w-_", exist_edge[ ,2], fixed=T)
		id_pos <- grep("w+_", exist_edge[ ,2], fixed=T)
	
		if (length(id_pos) > 0) {
			pos_edge <- gsub("w+_", "", exist_edge[id_pos,2], fixed=T)
			pos_edge <- strsplit(pos_edge, "_")
			for (i in 1:length(id_pos))
				adja[as.integer(pos_edge[[i]][1]), as.integer(pos_edge[[i]][2])] <- as.double(exist_edge[id_pos[i],1])
		}
	
		if (length(id_neg) > 0) {  
			neg_edge <- gsub("w-_", "", exist_edge[id_neg,2], fixed=T)
			neg_edge <- strsplit(neg_edge, "_")
		
			for (i in 1:length(id_neg)) {
				id1 <- as.integer(neg_edge[[i]][1])
				id2 <- as.integer(neg_edge[[i]][2])
				
				# check whether there is already a positive edge: if yes: take the one with the higher weight
				if (adja[id1,id2] != 0) {
					val_neg <- as.double(exist_edge[id_neg[i],1])
					
					if (adja[id1,id2] < val_neg)  # if negative is bigger: take negative
						adja[id1,id2] <- -val_neg
					
					if (adja[id1,id2] == val_neg)  # if both are equal: take average: 0
						adja[id1,id2] <- 0
					
					# if positive is bigger: take positive -> thus, do nothing
				}
				else { 
					adja[id1,id2] <- -as.double(exist_edge[id_neg[i],1])
				}
			}
		}
  }
  return(adja)
}