|
a |
|
b/weighted.statsOLD.R |
|
|
1 |
weighted.stats = |
|
|
2 |
function (x, w, c) |
|
|
3 |
{ |
|
|
4 |
n = ncol(x) #number of samples |
|
|
5 |
p = nrow(x) #number of genes |
|
|
6 |
k = length(unique(c)) #number of class |
|
|
7 |
c = as.integer(c) |
|
|
8 |
WM = WS = matrix(0, p, k) |
|
|
9 |
rownames(WS) = rownames(x) |
|
|
10 |
colnames(WS) = unique(c) |
|
|
11 |
c0 <- as.integer(min(c, na.rm = TRUE) - 1) |
|
|
12 |
c <- as.integer(c) - c0 |
|
|
13 |
mk = NULL |
|
|
14 |
|
|
|
15 |
w.mean00 = |
|
|
16 |
function (x, w) |
|
|
17 |
{ |
|
|
18 |
wm = NULL |
|
|
19 |
|
|
|
20 |
for (i in 1:p) |
|
|
21 |
{ |
|
|
22 |
wm0 = sum(w[i,]*x[i,]) / sum(w[i,]) |
|
|
23 |
wm = c(wm, wm0) |
|
|
24 |
} |
|
|
25 |
return(wm) |
|
|
26 |
} |
|
|
27 |
|
|
|
28 |
w.mean = |
|
|
29 |
function (x, w, c) |
|
|
30 |
{ |
|
|
31 |
for (j in 1:k) |
|
|
32 |
{ |
|
|
33 |
WM[,j] = w.mean00(x[,c == j], w[,c == j]) |
|
|
34 |
} |
|
|
35 |
return(WM) |
|
|
36 |
} |
|
|
37 |
|
|
|
38 |
w.sd = |
|
|
39 |
function (x, w, c) |
|
|
40 |
{ |
|
|
41 |
w.sd00 = |
|
|
42 |
function (x, w) |
|
|
43 |
{ |
|
|
44 |
ws = NULL |
|
|
45 |
|
|
|
46 |
w.sd0 = |
|
|
47 |
function (x, w) |
|
|
48 |
{ |
|
|
49 |
sumw = sum(w) |
|
|
50 |
sumw.sq = sum(w)^2 |
|
|
51 |
w.sq = sum(w^2) |
|
|
52 |
denom = sum(w * ((x - mean(x))^2)) |
|
|
53 |
sqrt((sumw * denom) / (sumw.sq - w.sq)) |
|
|
54 |
} |
|
|
55 |
|
|
|
56 |
for (i in 1:p) |
|
|
57 |
{ |
|
|
58 |
ws0 = w.sd0(x[i,], w[i,]) |
|
|
59 |
ws = c(ws, ws0) |
|
|
60 |
} |
|
|
61 |
|
|
|
62 |
return(ws) |
|
|
63 |
} |
|
|
64 |
|
|
|
65 |
for (j in 1:k) |
|
|
66 |
{ |
|
|
67 |
WS[,j] = w.sd00(x[,c == j], w[,c == j]) |
|
|
68 |
} |
|
|
69 |
return(WS) |
|
|
70 |
} |
|
|
71 |
|
|
|
72 |
WMEAN = w.mean00(x, w) #Overall weighted mean |
|
|
73 |
delta = WMEAN.G = w.mean(x, w, c) #Weighted means for each group |
|
|
74 |
WSD.G = w.sd(x, w, c) #Weighted standard deviations for each group |
|
|
75 |
WSD.POOLED = WSD.G |
|
|
76 |
|
|
|
77 |
for (i in 1:k) |
|
|
78 |
{ |
|
|
79 |
WSD.POOLED[,i] = (table(c)[i]-1) * (WSD.POOLED[,i]^2) |
|
|
80 |
mk[i] = sqrt((1 / table(c)[i]) + (1 / n)) |
|
|
81 |
} |
|
|
82 |
|
|
|
83 |
WSD.POOLED = sqrt(rowSums(as.data.frame(WSD.POOLED)) / (n - k)) |
|
|
84 |
s0 = median(WSD.POOLED) |
|
|
85 |
|
|
|
86 |
for (i in 1:k) |
|
|
87 |
{ |
|
|
88 |
delta[,i] = (delta[,i] - WMEAN) / (mk[i]*(WSD.POOLED + s0)) |
|
|
89 |
} |
|
|
90 |
|
|
|
91 |
rownames(WMEAN) = rownames(WMEAN.G) = rownames(delta) = rownames(WSD.G) = rownames(WSD.POOLED) = rownames(x) |
|
|
92 |
colnames(WMEAN.G) = colnames(delta) = colnames(WSD.G) = unique(c) |
|
|
93 |
|
|
|
94 |
stats = list(n = n, p = p, k = k, mk = mk, s0 = s0, delta = delta, WMEAN = WMEAN, WMEAN.G = WMEAN.G, WSD.G = WSD.G, WSD.POOLED = WSD.POOLED) |
|
|
95 |
|
|
|
96 |
return(stats) |
|
|
97 |
} |