In [1]:
library(qgraph)
library(igraph)
In [2]:
data_dir <- "ConD31"
theta <- 5
epsilon2 <- 0.01
In [3]:
#read data
freq <- as.matrix(read.csv(paste(data_dir,"/data/data_VAF.txt",sep=""), row.names=1, sep="\t"))
freq <- freq / 100
#add root data (freq=0.5)
freq <- rbind(freq, 0.5)
rownames(freq)[dim(freq)[1]] <- "root"
#remove low allele frequency data
for(i in dim(freq)[1]:1){
if(max(freq[i,]) < 0.05){
freq <- freq[-i,]
}
}
mnames <- rownames(freq)
mutation_num <- dim(freq)[1]
sample_num <- dim(freq)[2]
#Distance between mutations
Distance_matrix <- matrix(0, nrow=mutation_num, ncol=mutation_num)
for(i in 1:mutation_num){
for(j in 1:mutation_num){
Distance_matrix[i,j] <- sum((freq[i,]-freq[j,])**2) / (sample_num * 0.5 * (mean(freq[i,])+mean(freq[j,])))
}
}
diag(Distance_matrix) <- 0
minus_log_Distance_matrix <- matrix(nrow=mutation_num, ncol=mutation_num)
for(i in 1:mutation_num){
for(j in 1:mutation_num){
if(Distance_matrix[i,j] == 0){
minus_log_Distance_matrix[i,j] <- Inf
}
else{
minus_log_Distance_matrix[i,j] <- -log(Distance_matrix[i,j])
}
#remove edge that satisfy -log(d(i,j)) < theta
if(minus_log_Distance_matrix[i,j] < theta){
minus_log_Distance_matrix[i,j] <- 0
}
}
}
#clustering mutation into mutation nodes based on connected components of the above graph
g <- graph.adjacency(minus_log_Distance_matrix, weighted=TRUE, mode="undirected")
cls <- clusters(g, "weak")
mutation_node_num <- cls$no
mutation_node_list <- list()
for(i in 1:mutation_node_num){
mutation_node_list[[i]] <- which(cls$membership==i)
}
mutation_node_mnames <- rep("", mutation_node_num)
for(i in 1:mutation_node_num){
mutation_node_mnames[i] <- rownames(freq)[mutation_node_list[[i]]][[1]]
if(length(mutation_node_list[[i]]) >= 2){
for(j in 2:length(mutation_node_list[[i]])){
mutation_node_mnames[i] <- paste(mutation_node_mnames[i], rownames(freq)[mutation_node_list[[i]]][[j]], sep="\n")
}
}
}
#mean allele frequency of mutation node
mean_freq <- matrix(0, nrow=mutation_node_num, ncol=dim(freq)[2])
for(i in 1:mutation_node_num){
if(length(mutation_node_list[[i]]) > 1){
mean_freq[i,] <- colSums(freq[mutation_node_list[[i]],]) / length(mutation_node_list[[i]])
}
else{
mean_freq[i,] <- freq[mutation_node_list[[i]][1],]
}
}
#inference of one mother two daughter nodes relationships
epsilon1 <- 0.005
Issum_matrix <- matrix(rep(0,mutation_node_num*2), nrow=mutation_node_num, ncol=mutation_node_num)
while(epsilon1 > 0.0){
print(paste("current epsilon1 =",epsilon1,sep=" "))
Issum_matrix <- diag(rep(0,mutation_node_num))
for(i in 1:mutation_node_num){
for(j in 1:(mutation_node_num-1)){
for(k in (j+1):mutation_node_num){
if(i == j || i == k){
next
}
Y <- mean_freq[i,]
X <- mean_freq[j,] + mean_freq[k,]
tmp <- sum((Y-X)**2/sample_num)
tmp <- tmp / mean(X)
if(tmp < epsilon1){
Issum_matrix[j,i] <- 1
Issum_matrix[k,i] <- 1
}
}
}
}
if(sum(apply((Issum_matrix!=0), MARGIN=2, sum) > 2) > 0 || (sum(apply((Issum_matrix!=0), MARGIN=1, sum) > 1)) > 0){
epsilon1 <- epsilon1 - 0.00025
}
else{
break
}
}
jpeg(paste(data_dir,"/out/lineage_step1.jpeg",sep=""),height=960, width=960, res=144)
q1 <- qgraph(t(Issum_matrix!=0), layout="spring", labels=mutation_node_mnames, edge.color="red")
dev.off()
options(repr.plot.width=15, repr.plot.height=15)
plot(q1)
In [4]:
#additioanl lineage inference
Total_lineage_tree <- (Issum_matrix!=0)
leaf_nodes <- c()
orphaned_nodes <- c()
old_Total_lineage_tree <- Total_lineage_tree
mutation_node_mnames2 <- mutation_node_mnames
mutation_node_num2 <- mutation_node_num
mean_freq2 <- mean_freq
#find leaf nodes
root_id <- which(mutation_node_mnames=="root")
check_leaf_node <- function(i){
if(sum(Total_lineage_tree[,i]) > 0){
check_leaf_node(which(Total_lineage_tree[,i]==1)[1])
check_leaf_node(which(Total_lineage_tree[,i]==1)[2])
}
else{
leaf_nodes <<- c(leaf_nodes, i)
}
}
check_leaf_node(root_id)
#find orphaned nodes
for(i in 1:mutation_node_num){
if(i == root_id){
next
}
if(sum(Total_lineage_tree[i,]) == 0){
orphaned_nodes <<- c(orphaned_nodes, i)
}
}
#investigate orphaned nodes in allele frequency ascending order
tmp <- order(rowSums(mean_freq2[orphaned_nodes,]), decreasing=TRUE)
orphaned_nodes <- orphaned_nodes[tmp]
tmp_count <- 1
for(i in 1:length(orphaned_nodes)){
child_id <- orphaned_nodes[i]
tmp_parent <- rep(0, mutation_node_num2)
for(j in 1:length(leaf_nodes)){
parent_id <- leaf_nodes[j]
if(sum((mean_freq2[parent_id,]-mean_freq2[child_id,]+epsilon2)<0) == 0){
tmp_parent[parent_id] <- 1
}
}
#only if the orphaned node has one possible mother node
if(sum(tmp_parent) == 1){
parent_id <- which(tmp_parent==1)
Total_lineage_tree[child_id,parent_id] <- 1
#add pseudo node (push pseudo node for mean_freq and tree)
tmp_tree <- matrix(0, nrow=dim(Total_lineage_tree)[1]+1, ncol=dim(Total_lineage_tree)[2]+1)
tmp_tree[1:(dim(Total_lineage_tree)[1]),1:(dim(Total_lineage_tree)[2])] <- Total_lineage_tree
tmp_tree[dim(tmp_tree)[1],parent_id] <- 1
mutation_node_mnames2 <- c(mutation_node_mnames2, paste("Pseudo_",tmp_count,sep=""))
tmp_count <- tmp_count + 1
tmp_mean_freq <- matrix(0, nrow=dim(mean_freq2)[1]+1, ncol=dim(mean_freq2)[2])
tmp_mean_freq[1:dim(mean_freq2)[1],] <- mean_freq2
tmp_mean_freq[dim(mean_freq2)[1]+1,] <- mean_freq2[parent_id,] - mean_freq2[child_id,]
Total_lineage_tree <- tmp_tree
mean_freq2 <- tmp_mean_freq
mutation_node_num2 <- mutation_node_num2+1
#remake assigned leaf node
leaf_nodes <<- c()
check_leaf_node(root_id)
}
else if(sum(tmp_parent) > 1){
print(paste("Node :", mutation_node_mnames2[child_id], "cannot be assigned uniquely."))
for(j in 1:sum(tmp_parent)){
print(paste(" ", mutation_node_mnames2[child_id], " < ", mutation_node_mnames2[which(tmp_parent==1)[j]]))
}
}
else{
print(paste("Node :", mutation_node_mnames2[child_id], "cannot be assigned any leaf node."))
}
}
jpeg(paste(data_dir,"/out/lineage_step2.jpeg",sep=""),height=960, width=960, res=144)
q2 <- qgraph(t(Total_lineage_tree), layout="spring", labels=mutation_node_mnames2, edge.color="red")
dev.off()
plot(q2)
In [ ]: