forked from flatironinstitute/CaImAn-MATLAB
-
Notifications
You must be signed in to change notification settings - Fork 0
/
update_temporal_components.m
384 lines (364 loc) · 16.9 KB
/
update_temporal_components.m
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
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
function [C,f,P,S,YrA] = update_temporal_components(Y,A,b,Cin,fin,P,options)
% update temporal components and background given spatial components
% A variety of different methods can be used and are separated into 2 classes:
% 1-d approaches, where for each component a 1-d trace is computed by removing
% the effect of all the other components and then averaging with the corresponding
% spatial footprint. Then each trace is denoised. This corresponds to a block-coordinate approach
% 4 different 1-d approaches are included, and any custom method
% can be easily incorporated:
% 'project': The trace is projected to satisfy the constraints by the (known) calcium indicator dynamics
% 'constrained_foopsi': The noise constrained deconvolution approach is used. Time constants can be re-estimated (default)
% 'MCEM_foopsi': Alternating between constrained_foopsi and a MH approach for re-estimating the time constants
% 'MCMC': A fully Bayesian method (slowest, but usually most accurate)
% multi-dimensional approaches: (slowest)
% 'noise_constrained':
% C(j,:) = argmin_{c_j} sum(G*c_j),
% subject to: G*c_j >= 0
% ||Y(i,:) - A*C - b*f|| <= sn(i)*sqrt(T)
% The update can happen either in parallel (default) or serial by tuning options.temporal_parallel.
% In the case of parallel implementation the methods 'MCEM_foopsi' and 'noise_constrained' are not supported
% INPUTS:
% Y: raw data ( d X T matrix)
% A: spatial footprints (d x nr matrix)
% b: spatial background (d x 1 vector)
% Cin: current estimate of temporal components (nr X T matrix)
% fin: current estimate of temporal background (1 x T vector)
% P: struct for neuron parameters
% options: struct for algorithm parameters
% LD: Lagrange multipliers (needed only for 'noise_constrained' method).
%
% OUTPUTS:
% C: temporal components (nr X T matrix)
% f: temporal background (1 x T vector)
% P: struct for neuron parameters
% S: deconvolved activity
% Written by:
% Eftychios A. Pnevmatikakis, Simons Foundation, 2015
memmaped = isobject(Y);
if memmaped
sizY = size(Y,'Y');
d = prod(sizY(1:end-1));
T = sizY(end);
else
[d,T] = size(Y);
end
if isempty(P) || nargin < 6
active_pixels = find(sum(A,2)); % pixels where the greedy method found activity
unsaturated_pixels = find_unsaturatedPixels(Y); % pixels that do not exhibit saturation
options.pixels = intersect(active_pixels,unsaturated_pixels); % base estimates only on unsaturated, active pixels
end
defoptions = CNMFSetParms;
if nargin < 7 || isempty(options); options = []; end
if ~isfield(options,'deconv_method') || isempty(options.deconv_method); method = defoptions.deconv_method; else method = options.deconv_method; end % choose method
if ~isfield(options,'restimate_g') || isempty(options.restimate_g); restimate_g = defoptions.restimate_g; else restimate_g = options.restimate_g; end % re-estimate time constant (only with constrained foopsi)
if ~isfield(options,'temporal_iter') || isempty(options.temporal_iter); ITER = defoptions.temporal_iter; else ITER = options.temporal_iter; end % number of block-coordinate descent iterations
if ~isfield(options,'bas_nonneg'); options.bas_nonneg = defoptions.bas_nonneg; end
if ~isfield(options,'fudge_factor'); options.fudge_factor = defoptions.fudge_factor; end
if ~isfield(options,'temporal_parallel'); options.temporal_parallel = defoptions.temporal_parallel; end
if isfield(P,'interp'); Y_interp = P.interp; else Y_interp = sparse(d,T); end % missing data
if isfield(P,'unsaturatedPix'); unsaturatedPix = P.unsaturatedPix; else unsaturatedPix = 1:d; end % saturated pixels
mis_data = find(Y_interp); % interpolate any missing data before deconvolution
if ~memmaped
Y(mis_data) = Y_interp(mis_data);
end
if (strcmpi(method,'noise_constrained') || strcmpi(method,'project')) && ~isfield(P,'g')
options.flag_g = 1;
if ~isfield(P,'p') || isempty(P.p); P.p = 2; end;
p = P.p;
P = arpfit(Yr,p,options,P.sn);
if ~iscell(P.g)
G = make_G_matrix(T,P.g);
end
else
G = speye(T);
end
ff = find(sum(A)==0);
if ~isempty(ff)
A(:,ff) = [];
if exist('Cin','var')
if ~isempty(Cin)
Cin(ff,:) = [];
end
end
end
% estimate temporal (and spatial) background if they are not present
if isempty(fin) || nargin < 5 % temporal background missing
bk_pix = (sum(A,2)==0); % pixels with no active neurons
if isempty(b) || nargin < 3
fin = mean(Y(bk_pix,:));
fin = fin/norm(fin);
b = max(Y*fin',0);
else
fin = max(b(bk_pix)'*Y(bk_pix,:),0)/norm(b(bk_pix))^2;
end
end
if isempty(Cin) || nargin < 4 % estimate temporal components if missing
Cin = max((A'*A)\(A'*Y - (A'*b)*fin),0);
ITER = max(ITER,3);
end
if isempty(b) || isempty(fin) || nargin < 5 % re-estimate temporal background
if isempty(b) || nargin < 3
[b,fin] = nnmf(max(Y - A*Cin,0),1);
else
fin = max((b'*Y - (b'*A)*Cin)/norm(b)^2,0);
end
end
if ~memmaped
saturatedPix = setdiff(1:d,unsaturatedPix); % remove any saturated pixels
Ysat = Y(saturatedPix,:);
Asat = A(saturatedPix,:);
bsat = b(saturatedPix,:);
Y = Y(unsaturatedPix,:);
A = A(unsaturatedPix,:);
b = b(unsaturatedPix,:);
d = length(unsaturatedPix);
end
K = size(A,2);
A = [A,b];
S = zeros(size(Cin));
Cin = [Cin;fin];
C = Cin;
if strcmpi(method,'noise_constrained')
Y_res = Y - A*Cin;
mc = min(d,15); % number of constraints to be considered
LD = 10*ones(mc,K);
else
step = 5e3;
nA = sum(A.^2);
AA = (A'*A)/spdiags(nA(:),0,length(nA),length(nA));
if memmaped
YA = zeros(T,length(nA));
for i = 1:step:d
YA = YA + double(Y.Yr(i:min(i+step-1,d),:))'*A(i:min(i+step-1,d),:);
end
YA = YA/spdiags(nA(:),0,length(nA),length(nA));
else
YA = (Y'*A)/spdiags(nA(:),0,length(nA),length(nA));
end
YrA = (YA - Cin'*AA);
if strcmpi(method,'constrained_foopsi') || strcmpi(method,'MCEM_foopsi')
P.gn = cell(K,1);
P.b = cell(K,1);
P.c1 = cell(K,1);
P.neuron_sn = cell(K,1);
end
if strcmpi(method,'MCMC')
params.B = 300;
params.Nsamples = 400;
params.p = P.p;
else
params = [];
end
end
p = P.p;
if options.temporal_parallel
for iter = 1:ITER
[O,lo] = update_order(A(:,1:K));
for jo = 1:length(O)
Ytemp = YrA(:,O{jo}(:)) + Cin(O{jo},:)';
Ctemp = zeros(length(O{jo}),T);
Stemp = zeros(length(O{jo}),T);
btemp = zeros(length(O{jo}),1);
sntemp = btemp;
c1temp = btemp;
gtemp = cell(length(O{jo}),1);
% FN added the part below in order to save SAMPLES as a field of P
if strcmpi(method,'MCMC')
clear samples_mcmc
samples_mcmc(length(O{jo})) = struct();
[samples_mcmc.Cb] = deal(zeros(params.Nsamples,1));
[samples_mcmc.Cin] = deal(zeros(params.Nsamples,1));
[samples_mcmc.sn2] = deal(zeros(params.Nsamples,1));
[samples_mcmc.ns] = deal(zeros(params.Nsamples,1));
[samples_mcmc.ss] = deal(cell(params.Nsamples,1));
[samples_mcmc.ld] = deal(zeros(params.Nsamples,1));
[samples_mcmc.Am] = deal(zeros(params.Nsamples,1));
[samples_mcmc.g] = deal(zeros(params.Nsamples,1));
[samples_mcmc.params] = deal(struct('lam_', [], 'spiketimes_', [], 'A_', [], 'b_', [], 'C_in', [], 'sg', [], 'g', []));
end
parfor jj = 1:length(O{jo})
if p == 0 % p = 0 (no dynamics assumed)
cc = max(Ytemp(:,jj),0);
Ctemp(jj,:) = full(cc');
Stemp(jj,:) = Ctemp(jj,:);
else
switch method
case 'project'
cc = plain_foopsi(Ytemp(:,jj),G);
Ctemp(jj,:) = full(cc');
Stemp(jj,:) = Ctemp(jj,:)*G';
case 'constrained_foopsi'
%if restimate_g
[cc,cb,c1,gn,sn,spk] = constrained_foopsi(Ytemp(:,jj),[],[],[],[],options);
%else
% [cc,cb,c1,gn,sn,spk] = constrained_foopsi(Ytemp(:,jj)/nA(jj),[],[],P.g,[],options);
%end
gd = max(roots([1,-gn'])); % decay time constant for initial concentration
gd_vec = gd.^((0:T-1));
Ctemp(jj,:) = full(cc(:)' + cb + c1*gd_vec);
Stemp(jj,:) = spk(:)';
Ytemp(:,jj) = Ytemp(:,jj) - Ctemp(jj,:)';
btemp(jj) = cb;
c1temp(jj) = c1;
sntemp(jj) = sn;
gtemp{jj} = gn(:)';
case 'MCMC'
SAMPLES = cont_ca_sampler(Ytemp(:,jj),params);
Ctemp(jj,:) = make_mean_sample(SAMPLES,Ytemp(:,jj));
Stemp(jj,:) = mean(samples_cell2mat(SAMPLES.ss,T));
btemp(jj) = mean(SAMPLES.Cb);
c1temp(jj) = mean(SAMPLES.Cin);
sntemp(jj) = sqrt(mean(SAMPLES.sn2));
gtemp{jj} = mean(exp(-1./SAMPLES.g))';
samples_mcmc(jj) = SAMPLES; % FN added.
end
end
end
if p > 0
if strcmpi(method,'constrained_foopsi') || strcmpi(method,'MCMC');
P.b(O{jo}) = num2cell(btemp);
P.c1(O{jo}) = num2cell(c1temp);
P.neuron_sn(O{jo}) = num2cell(sntemp);
for jj = 1:length(O{jo})
P.gn(O{jo}(jj)) = gtemp(jj);
end
YrA = YrA - (Ctemp-C(O{jo}(:),:))'*AA(O{jo}(:),:);
C(O{jo}(:),:) = Ctemp;
S(O{jo}(:),:) = Stemp;
if strcmpi(method,'MCMC');
P.samples_mcmc(O{jo}) = samples_mcmc; % FN added, a useful parameter to have.
end
end
else
YrA = YrA - (Ctemp-C(O{jo}(:),:))'*AA(O{jo}(:),:);
C(O{jo}(:),:) = Ctemp;
S(O{jo}(:),:) = Stemp;
%YrA = (YA - C'*AA)/spdiags(nA(:),0,length(nA),length(nA));
end
fprintf('%i out of %i components updated \n',sum(lo(1:jo)),K);
end
for ii = K + 1:size(C,1)
%YrA(:,ii) = YrA(:,ii) + Cin(ii,:)';
cc = full(max(YrA(:,ii)'+Cin(ii,:),0));
YrA = YrA - (cc-C(ii,:))'*AA(ii,:);
C(ii,:) = cc; %full(cc');
%YrA(:,ii) = YrA(:,ii) - C(ii,:)';
%YrA(:,end) = (YA(:,end) - C'*AA(:,end)); %/nA(end); %spdiags(nA(:),0,length(nA),length(nA));
end
%YrA = (YA - Cin'*AA)/spdiags(nA(:),0,length(nA),length(nA));
if norm(Cin - C,'fro')/norm(C,'fro') <= 1e-3
% stop if the overall temporal component does not change by much
break;
else
Cin = C;
end
end
else
for iter = 1:ITER
perm = randperm(K+size(b,2));
for jj = 1:K
ii = perm(jj);
if ii<=K
ff = find(AA(ii,:));
if P.p == 0 % p = 0 (no dynamics assumed)
%YrA(:,ii) = YrA(:,ii) + Cin(ii,:)';
Ytemp = YrA(:,ii) + Cin(ii,:)';
cc = max(Ytemp,0);
YrA(:,ff) = YrA(:,ff) - (cc - C(ii,:)')*AA(ii,ff);
C(ii,:) = full(cc');
S(ii,:) = C(ii,:);
else
switch method
case 'project'
%YrA(:,ii) = YrA(:,ii) + Cin(ii,:)';
Ytemp = YrA(:,ii) + Cin(ii,:)';
cc = plain_foopsi(Ytemp,G);
YrA(:,ff) = YrA(:,ff) - (cc - C(ii,:)')*AA(ii,ff);
C(ii,:) = full(cc');
S(ii,:) = C(ii,:)*G';
case 'constrained_foopsi'
%YrA(:,ii) = YrA(:,ii) + Cin(ii,:)';
Ytemp = YrA(:,ii) + Cin(ii,:)';
if restimate_g
[cc,cb,c1,gn,sn,spk] = constrained_foopsi(Ytemp,[],[],[],[],options);
P.gn{ii} = gn;
else
[cc,cb,c1,gn,sn,spk] = constrained_foopsi(Ytemp,[],[],P.g,[],options);
end
gd = max(roots([1,-gn'])); % decay time constant for initial concentration
gd_vec = gd.^((0:T-1));
YrA(:,ff) = YrA(:,ff) - (cc(:) + cb + c1*gd_vec' - C(ii,:)')*AA(ii,ff);
C(ii,:) = full(cc(:)' + cb + c1*gd_vec);
S(ii,:) = spk(:)';
%YrA(:,ii) = YrA(:,ii) - C(ii,:)';
P.b{ii} = cb;
P.c1{ii} = c1;
P.neuron_sn{ii} = sn;
case 'MCEM_foopsi'
options.p = length(P.g);
%YrA(:,ii) = YrA(:,ii) + Cin(ii,:)';
Ytemp = YrA(:,ii) + Cin(ii,:)';
[cc,cb,c1,gn,sn,spk] = MCEM_foopsi(Ytemp,[],[],P.g,[],options);
gd = max(roots([1,-gn.g(:)']));
gd_vec = gd.^((0:T-1));
YrA(:,ff) = YrA(:,ff) - (cc(:) + cb + c1*gd_vec' - C(ii,:)')*AA(ii,ff);
C(ii,:) = full(cc(:)' + cb + c1*gd_vec);
S(ii,:) = spk(:)';
%YrA(:,ii) = YrA(:,ii) - C(ii,:)';
P.b{ii} = cb;
P.c1{ii} = c1;
P.neuron_sn{ii} = sn;
P.gn{ii} = gn.g;
case 'MCMC'
params.B = 300;
params.Nsamples = 400;
params.p = P.p; %length(P.g);
%YrA(:,ii) = YrA(:,ii) + Cin(ii,:)';
Ytemp = YrA(:,ii) + Cin(ii,:)';
SAMPLES = cont_ca_sampler(Ytemp,params);
ctemp = make_mean_sample(SAMPLES,Ytemp);
YrA(:,ff) = YrA(:,ff) - (ctemp - C(ii,:))'*AA(ii,ff);
C(ii,:) = ctemp';
S(ii,:) = mean(samples_cell2mat(SAMPLES.ss,T));
P.b{ii} = mean(SAMPLES.Cb);
P.c1{ii} = mean(SAMPLES.Cin);
P.neuron_sn{ii} = sqrt(mean(SAMPLES.sn2));
gr = mean(exp(-1./SAMPLES.g));
gp = poly(gr);
P.gn{ii} = -gp(2:end);
P.samples_mcmc(ii) = SAMPLES; % FN added, a useful parameter to have.
case 'noise_constrained'
Y_res = Y_res + A(:,ii)*Cin(ii,:);
[~,srt] = sort(A(:,ii),'descend');
ff = srt(1:mc);
[cc,LD(:,ii)] = lagrangian_foopsi_temporal(Y_res(ff,:),A(ff,ii),T*P.sn(unsaturatedPix(ff)).^2,G,LD(:,ii));
C(ii,:) = full(cc');
Y_res = Y_res - A(:,ii)*cc';
S(ii,:) = C(ii,:)*G';
end
end
else
%YrA(:,ii) = YrA(:,ii) + Cin(ii,:)';
%cc = max(YrA(:,ii),0);
Ytemp = YrA(:,ii) + Cin(ii,:)';
cc = max(Ytemp,0);
%YrA(:,ii) = YrA(:,ii) - (cc-C(ii,:))'*AA(ii,:);
YrA = YrA - (cc - C(ii,:)')*AA(ii,:);
C(ii,:) = full(cc');
%YrA(:,ii) = YrA(:,ii) - C(ii,:)';
end
if mod(jj,10) == 0
fprintf('%i out of total %i temporal components updated \n',jj,K);
end
end
if norm(Cin - C,'fro')/norm(C,'fro') <= 1e-3
% stop if the overall temporal component does not change by much
break;
else
Cin = C;
end
end
end
f = C(K+1:end,:);
C = C(1:K,:);
YrA = YrA(:,1:K)';