TreeKernel/hcpcSims.R at main · Ghoshlab/TreeKernel · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
###############################
##
## Project: MetaboGuru
##
## Purpose: Messing with hierarchical Simulations
##
## Author: Charlie Carpenter
## Email: charles.carpenter@cuanschutz.edu
##
## Date Created: 2022-05-23
##
## ---------------------------
## Notes:
##
##
## ---------------------------

## Helpful functions
source('~/Documents/Research/Current/MetaboGuru/Carpenter/RCode/ScoreSimFunctions.R')

# HCTree ------------------------------------------------------------------

## Function to simulate higher order interactions
## between covariate and kernel space
## From https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6415770/
## graph for Z, Covariates, depth of kdTree, a numerical vector (length=2),
## sample size, size of graph, covariate beta vector,
## standard deviation of y
hcTreeIntDat <- function(graph, covs, zz, groups=2,
                         n, p, b0, sd.y, classifier='continuous'){

  stopifnot("All length(zz) must equal p"= all(sapply(zz,length)==p))
  stopifnot("length(zz) must equal groups"= length(zz)==groups)

  if(classifier=='continuous'){
    pca <- PCA(covs, graph=F)
    hcpcFull <- HCPC(pca, nb.clust=groups, min=groups, graph=F)
  }
  if(classifier=='discrete'){
    famd <- FAMD(covs, graph=F)
    hcpcFull <- HCPC(famd, nb.clust=4, min=4, graph=F)
  }

  tr <- hcpcFull$call$t$tree
  hcpc <- hcpcFull$data.clust
  hcpc$clust <- as.numeric(hcpc$clust)

  A <- as.matrix(get.adjacency(graph))
  test <- A + diag(p)
  Omega1 <- Danaher_pos_def(test)
  Z <- mvrnorm(n=nrow(covs), rep(0,p), solve(Omega1))

  ## Partitions
  if(any(hcpc$clust>groups)){
    hcpc$clust <- ifelse(hcpc$clust>groups,groups,hcpc$clust)
  }

  mX <- model.matrix(~X1+X2+X3+X4, data=covs)
  null <- mX%*%rep(b0, ncol(mX))
  Y <- VY <- ZK <- numeric(n)

  for(i in 1:length(unique(hcpc$clust)) ){
    kk <- hcpc$clust==i
    Zk <- Z[kk,] %*% zz[[i]]
    ZK[kk] <- Zk
    Y[kk] <- null[kk] + Zk
  }

  Ve <- rnorm(n=n, sd=sd.y)
  Y <- Y + Ve

  list(Y=Y, graph=graph, covs=covs, Z=Z, tr=tr, Ve=var(Ve), VY=var(ZK))
}


## Function to perform TreeKernel on output from hcTreeIntDat
## This is the function used for simulations
hcTreeTest <- function(hcDat, formula.H0, groups=2, classifier='continuous', ncp=5){

  Y <- hcDat$Y; covs <- hcDat$covs
  graph <- hcDat$graph; Z <- hcDat$Z
  dd <- data.frame(Y, covs)
  mod <- lm(formula.H0, data=dd)
  R <- mod$residuals
  dd <- data.frame(R,dd)

  I <- diag(length(V(graph)))
  NL <- as.matrix( graph.laplacian(graph, normalized = T) )
  RNL <- solve(I + tau*NL)

  if(classifier=='continuous'){
    pca <- PCA(covs, graph=F, ncp=ncp)
    hcpc <- HCPC(pca, nb.clust = groups, min=groups, graph=F)$data.clust
  }
  if(classifier=='discrete'){
    famd <- FAMD(covs, graph=F, ncp=ncp)
    hcpc <- HCPC(famd, nb.clust = groups, min=groups, graph=F)$data.clust
  }

  dd <- data.frame(dd, hcpc)

  ## testing within each partition
  pval <- numeric()
  for(i in 1:length(unique(hcpc$clust)) ){
    kk <- dd$clust==i
    if(sum(kk)==1) next
    ds <- dd[kk,]
    Zs <- scale(Z[kk, ,drop=F])
    rho <- median(dist(Zs))
    Zl <- try(Zs %*% RNL, silent = T)

    Ks <- Gaussian_kernel(rho, Zl)

    pval[i] <- SKAT.c(R~1, data=ds, K=Ks)$Qq
  }

  Zs <- scale(Z)
  rho <- median(dist(Zs))
  Zl <- Zs %*% RNL
  K <- Gaussian_kernel(rho, Zl)

  fullP <- SKAT.c(formula.H0, data=dd, K=K)$Qq

  list(pvalHC = pval, fullP=fullP, tr = hcDat$tr)
}


hcTreeIntDatOneTree <- function(graph, covs, hcpcFull, zz, groups=2,
                                n, p, b0, sd.y){

  stopifnot("All length(zz) must equal p"= all(sapply(zz,length)==p))
  stopifnot("length(zz) must equal groups"= length(zz)==groups)

  tr <- hcpcFull$call$t$tree
  hcpc <- hcpcFull$data.clust
  hcpc$clust <- as.numeric(hcpc$clust)

  A <- as.matrix(get.adjacency(graph))
  test <- A + diag(p)
  Omega1 <- Danaher_pos_def(test)
  Z <- mvrnorm(n=nrow(covs), rep(0,p), solve(Omega1))

  ## Partitions
  if(any(hcpc$clust>groups)){
    hcpc$clust <- ifelse(hcpc$clust>groups,groups,hcpc$clust)
  }

  mX <- as.matrix(cbind(1, covs))
  null <- mX%*%b0
  Y <- VY <- ZK <- numeric(n)
  for(i in 1:groups){
    kk <- hcpc$clust==i
    Zk <- Z[kk,] %*% zz[[i]]

    ZK[kk] <- Zk
    Y[kk] <- null[kk] + Zk
  }

  Ve <- rnorm(n=n, sd=sd.y)
  Y <- Y + Ve

  list(Y=Y, graph=graph, covs=covs, Z=Z, tr=tr, Ve=var(Ve), VY=var(ZK))
}