##GeneExp: the Gene (mRNAs, TFs) Expression Data (the rows are Genes and the colomns are samples)
##miRNAExp: the miRNA Expression Data(the rows are miRNAs and the colomns are samples)
## miRNATFmRNA_network: the miRNA-TF-mRNA regulatory network [row(regulators)->colomn(tagagets)]. 
##                      The number of row (column) is the sum of the genes (mRNAs, TFs) and miRNAs.
##                      The order of features in row (column) is corresponding to the order 
##                      in Gene Expression data and miRNA expression data. 
##beta: the tuning parameter for the importance of the miRNA-TF-mRNA regulatory network information.
##clusterNum: the clustering number for cancer subtypes

WSNF<-function(GeneExp,miRNAExp,miRNATFmRNA_network,beta=0.8,clusterNum=3) 
{
  library("SNFtool")
  ####1. Modified google PageRank algorithm
  # To excute the PageRank algorithm with a high-dimensions netwok matrix in R is a 
  # time-consuming work. However, this can be easily achieved in Matlab because Matlab has 
  # a strong processing capacity for sparse matrix. 
  # If you can't run this function in your computer, We recormmend to caculate the ranking of
  # PageRank in Matlab. Then use the "WSNF2" fuction below to get the result of cancer subtypes. 
  # We have also provied the calculated result of the ranking of each feature. It can be used directly. 
  colsum=colSums(miRNATFmRNA_network)
  colsum[which(colsum==0)]=1
  dev=t(apply(miRNATFmRNA_network, 1, function(x) x/colsum))
  num=nrow(dev)
  damping=0.85
  A=diag(num)-damping*dev
  B=solve(A)
  e=rep(1,num)
  delta=(1-damping)/num
  ranking=B %*% e * delta
  ranking=ranking/sum(ranking)
  
  ####2. Normalise ranking and MAD
  ##### a.Normalise ranking
  Gene_pageRank=ranking[1:nrow(GeneExp)]
  Gene_pageRank=Gene_pageRank/sum(Gene_pageRank)
  miRNA_pageRank=ranking[(nrow(GeneExp)+1):length(ranking)]
  miRNA_pageRank=miRNA_pageRank/sum(miRNA_pageRank)
  ##### b.Normalise MAD
  Gene_MAD=as.vector(apply(GeneExp,1,mad))
  Gene_MAD=Gene_MAD/sum(Gene_MAD)
  miRNA_MAD=as.vector(apply(miRNAExp,1,mad))
  miRNA_MAD=miRNA_MAD/sum(miRNA_MAD)
  ##### c.combine ranking and MAD
  Gene_combineweight = Gene_pageRank*beta+Gene_MAD*(1-beta)
  miRNA_combineweight = miRNA_pageRank*beta+miRNA_MAD*(1-beta)
  
  ####3.Weighted similartiy fusion network
  ###weighted distance fuction
  distanceWeighted<-function(X,weight)
  ##X is the expression Matrix(Row is featrue, column is sample) 
  {
    X_row = nrow(X)      
    weight_diag<-diag(weight)
    X2<-(X^2)%*%weight_diag 
    sumsqX = rowSums(X2)
    X1<-X%*%weight_diag
    XY = X1 %*% t(X)
    XX=matrix(rep(sumsqX, times = X_row), X_row, X_row)
    res=XX+t(XX)-2*XY
    res[res < 0] = 0
    diag(res)=0
    res<-sqrt(res)
    return(res)
  }
  
  Dist1=distanceWeighted(as.matrix(t(GeneExp)),Gene_combineweight)
  W1= affinityMatrix(Dist1, 20, 0.5)
  Dist2=distanceWeighted(as.matrix(t(miRNAExp)),miRNA_combineweight)
  W2=affinityMatrix(Dist2, 20, 0.5)
  W = SNF(list(W1,W2), 20, 20)
  group = spectralClustering(W,clusterNum)
  result=list(W=W,group=group,ranking=ranking)
  result  
}
##BRCA
result=WSNF(BRCA_gene,BRCA_miRNA,BRCA_miRNA_TF_mRNA_network,beta=0.8,clusterNum=5)
##GBM
result=WSNF(GBM_gene,GBM_miRNA,GBM_miRNA_TF_mRNA_network,beta=0.8,clusterNum=3)


####################################################################################

##GeneExp: the Gene (mRNAs, TFs) Expression Data (the rows are Genes and the colomns are samples)
##miRNAExp: the miRNA Expression Data (the rows are miRNAs and the colomns are samples)
##ranking: the ranking of each feature. The order of the ranking is corresponding to 
##         the order in Gene Expression data and miRNA expression data.
##beta: the tuning parameter for the importance of the miRNA-TF-mRNA regulatory network information.
##clusterNum: the clustering number for cancer subtypes
WSNF2<-function(GeneExp,miRNAExp,ranking,beta=0.8,clusterNum=3) 
{
  library("SNFtool")
  #### Normalise
  ##### a.Normalise ranking
  Gene_pageRank=ranking[1:nrow(GeneExp)]
  Gene_pageRank=Gene_pageRank/sum(Gene_pageRank)
  miRNA_pageRank=ranking[(nrow(GeneExp)+1):length(ranking)]
  miRNA_pageRank=miRNA_pageRank/sum(miRNA_pageRank)
  
  ##### b.Normalise MAD
  Gene_MAD=as.vector(apply(GeneExp,1,mad))
  Gene_MAD=Gene_MAD/sum(Gene_MAD)
  miRNA_MAD=as.vector(apply(miRNAExp,1,mad))
  miRNA_MAD=miRNA_MAD/sum(miRNA_MAD)
  ##### c.combine ranking and MAD
  Gene_combineweight=Gene_pageRank*beta+Gene_MAD*(1-beta)
  miRNA_combineweight=miRNA_pageRank*beta+miRNA_MAD*(1-beta)
  
  ####Weighted similartiy fusion network
  ###weighted distance fuction
  distanceWeighted<-function(X,weight)
    ##X is the expression Matrix(Row is featrue, column is sample) 
  {
    X_row = nrow(X)      
    weight_diag<-diag(weight)
    X2<-(X^2)%*%weight_diag 
    sumsqX = rowSums(X2)
    X1<-X%*%weight_diag
    XY = X1 %*% t(X)
    XX=matrix(rep(sumsqX, times = X_row), X_row, X_row)
    res=XX+t(XX)-2*XY
    res[res < 0] = 0
    diag(res)=0
    res<-sqrt(res)
    return(res)
  }
  
  Dist1=distanceWeighted(as.matrix(t(GeneExp)),Gene_combineweight)
  W1= affinityMatrix(Dist1, 20, 0.5)
  Dist2=distanceWeighted(as.matrix(t(miRNAExp)),miRNA_combineweight)
  W2=affinityMatrix(Dist2, 20, 0.5)
  W = SNF(list(W1,W2), 20, 20)
  group = spectralClustering(W,clusterNum)
  result=list(W=W,group=group,ranking=ranking)
  result  
}

##BRCA
result=WSNF2(BRCA_gene,BRCA_miRNA,BRCA_ranking[,3],beta=0.8,clusterNum=5)
##GBM
result=WSNF2(GBM_gene,GBM_miRNA,GBM_ranking[,3],beta=0.8,clusterNum=3)


############################################################################
# %%% This is Matlab code
# %%% Calculating the ranking of features in the miRNA-TF-mRNA regulatory network
# function ranking = FeatureRanking(Network);
# %%% Network: the miRNA-TF-mRNA regulatory network[row(regulators)->colomn(tagagets)]
# dumping=0.85;
# [Num,aa]=size(Network);
# delta=(1-dumping)/Num;
# s=sum(Network);
# index=find(s==0);
# s(index)=1;
# Deg=spdiags(1./s',0,Num,Num);
# e =ones(Num,1);
# w=(sparse(eye(Num))-dumping*Network*Deg)\e*delta;
# ranking=w/sum(w);