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
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
|
function res=mrses_hw_debug(A,B,k,Niter,Ncycle,distmod,block)
if (nargin<7)
block=256;
end
if (nargin<6)
distmod=1;
end
if (nargin<5)
Ncycle=1000;
end
if (nargin<4)
Niter=500;
end
if (nargin<3)
k=5;
end
if (nargin<2)
error('As minimum two matrixes needed for MRSES');
end
if (nargin>6)
error('Too much parameters');
end
sa=size(A);sb=size(B);
if (sa(2)==sb(2))
genes=sa(2);
else
error('Features dimension mismatch');
end
nA=sa(1); nB=sb(1);
%optki=zeros(Ncycle,k);
ctx = mrses_hw();
mrses_hw(ctx, 1, k, block, single(A), single(B), distmod);
mean_a = mean(A,1);
mean_b = mean(B,1);
mdiff = mean_a - mean_b;
dA = (A - ones([nA,1]) * mean_a) / sqrt(nA);
dB = (B - ones([nB,1]) * mean_b) / sqrt(nB);
for icycle=0:block:(Ncycle-1)
block_size = min(block, Ncycle - icycle);
% SELECT k GENES {ki} FOR TEST AND EXCLUDE THEM FROM ALL GENES {ke}
ki=int16([]); ke=int16([]);
for i=1:block_size
tt=randperm(genes);% randomizing genes
ki(:,i)=tt(1:k); % selecting first k
ke(:,i)=tt(k+1:end); % the rest unuzed
end
cur_dist_hw = mrses_hw(ctx, 10, block_size, ki);
cur_dist = multi_bmc(dA, dB, mdiff, ki, distmod);
% find(cur_dist ~= cur_dist_hw)
dist_diff = abs(cur_dist_hw - cur_dist);
allowed = abs(cur_dist)/100000;
find ( dist_diff > allowed)
% for i=1:block_size
% check_dist(i) = bmc(A(:,ki(:,i)),B(:,ki(:,i)), distmod);
% end
% find(cur_dist ~= check_dist)
for iter=1:Niter
xki=ceil(rand(1,block_size)*k); % selecting random gen from selected
xke=ceil(rand(1,block_size)*(genes-k)); % selected random gen from non-selected
idx_i = sub2ind(size(ki), xki, 1:block_size);
idx_e = sub2ind(size(ke), xke, 1:block_size);
t=ki(idx_i);
ki(idx_i)=ke(idx_e);
ke(idx_e)=t;
dist = multi_bmc(dA, dB, mdiff, ki, distmod);
% for i=1:block_size
% check_dist(i)=bmc(A(:,ki(:,i)),B(:,ki(:,i)),distmod); % compute distance between A and B with currently selected genes
% end
%find(dist ~= check_dist)
% dist_diff = abs(dist - check_dist);
% allowed = abs(dist)/1000000;
% find ( dist_diff > allowed)
bad = find(dist < cur_dist);
idx_i = idx_i(bad);
idx_e = idx_e(bad);
t=ki(idx_i);
ki(idx_i)=ke(idx_e);
ke(idx_e)=t;
cur_dist = max(dist, cur_dist);
end
optki(:,(icycle+1):(icycle+block_size))=ki; % save finally selected genes
end
mrses_hw(ctx);
optki=reshape(optki,1,[]);
[n,g]=hist(optki,1:genes);
H=[n./Ncycle;g];
res=flipud(sortrows(H'));
% DISTANCE CALCULATOR
% ifs are taking to much time, do a separate functions for each distance
function dist=multi_bmc(dA, dB, mean_diff, ki, distmod)
block_size = size(ki,2);
%ki(:,1) = [5,7,9,11,13];
for i=1:block_size
x = dA(:, ki(:,i));
%x(1:5,:)'
c1 = x'*x;
x = dB(:, ki(:,i));
c2 = x'*x;
c=(c1+c2)./2;
[L,p] = chol(c);
%detc = prod(diag(L,0))^2;
%tmp = diag(L,0);
%detc = tmp' * tmp;
if p > 0
detc = 0;
else
detc = det(L)^2;
end
% rcorr(i)=log((detc.*detc)./(det(c1).*det(c2)));
rcorr(i)=2.*log(detc./sqrt(det(c1).*det(c2)));
% rcorr(i)=2.*log(detc./sqrt(det(c1*c2)));
if detc == 0
mdiff = mean_diff(ki(:,i));
rmahal(i)=(mdiff*pinv(c))*mdiff';
else
% rmahal(i)=(mdiff/c)*mdiff';
% rmahal(i)=((mdiff/L)/L')*mdiff';
% rmahal(i) = prod(mdiff/L)^2;
tmp = mean_diff(ki(:,i))/L;
rmahal(i) = tmp * tmp';
end
end
if (distmod==1)
dist = rmahal./8 + rcorr./4;
elseif (distmode==2)
dist = rmahal;
else
dist = rcorr;
end
function dist=bmc(x1,x2,distmod)
c1=cov1(x1);
c2=cov1(x2);
c=(c1+c2)./2;
if (distmod~=2)
rcorr=2.*log(det(c)./sqrt(det(c1).*det(c2)));
end
if (distmod~=3)
m1=mean(x1);
m2=mean(x2);
rmahal=((m2-m1)/c)*(m2-m1)';
end
if (distmod==1)
dist = rmahal./8+rcorr./4;
elseif (distmode==2)
dist=rmahal;
else
dist=rcorr;
end
function c = cov1(x, m)
[rows, cols] = size(x);
%rows = size(x(:,1))
if (nargin<2)
nX = x - ones([rows,1]) * mean(x);
else
nX = x - ones([rows,1]) * m';
end
c = nX' * nX / rows;
function c = cov2(x)
c = x' * x;
|