#SOSCC algorithm
#George M. Garrity and Timothy G. Lilburn
#Copyright Michigan State University 2004, all rights reserved

"step8" <- function(taxon.subset, dis.table, all.genus, taxon.seq.new)
{
	all.genus <- substring(as.character(taxon.subset[, 4]), 1, regexpr("[^a-zA-Z]", as.character(taxon.subset[, 4])) - 1)
	taxon.level <- unique(all.genus)
	taxon.seq.refs <- taxon.seq.new

	outliers <- vector("list", length(taxon.level))
	names(outliers) <- taxon.level

	outlier <- 0
	for(i in 1:length(taxon.level)) {
		if(length(taxon.seq.new[[i]]) < 4) {
		}
		if(length(taxon.seq.new[[i]]) >= 4) {
			#estimate sample statistics
			temp.ref <- dis.table[taxon.seq.new[[i]], taxon.seq.new[[i]]]
			temp <- dis.table[taxon.seq.new[[i]], taxon.seq.new[[i]]]

			#identify outliers (> 3 stdev)
			outlier <- temp[temp < stdev(as.matrix(na.omit(temp.ref[lower.tri(temp.ref), diag = T]))) * 4]
			temp2 <- apply(temp, 1, FUN=function(x, y) match(as.vector(x), y), y = outlier)
			temp2 <- is.na(as.data.frame(temp2[, 1:nrow(temp2)])) * 1
			temp2 <- (apply(temp2, 1, sum))
			names(temp2) <- dimnames(temp)[[1]]
			outliers[[i]] <- as.character(names(temp2[temp2 > summary(temp2)[[5]]]))
		}
	}
	return(outliers)
}