-
Notifications
You must be signed in to change notification settings - Fork 13
/
glmfun_with_indicator.m
199 lines (177 loc) · 8.74 KB
/
glmfun_with_indicator.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
function [XX,P,I] = glmfun_with_indicator(Vlo_pre,Vlo_post,Vhi_pre,Vhi_post,pval,ci,varargin)
% a modification of glmfun, allowing the user to include an indicator
% function to the model, differentiating between a 'pre' and 'post'
% condition (e.g. before and after stimulus, with or without drug), to test
% for the effect of condition on PAC
%
% INPUTS:
% Vlo_pre: Low frequency signal from 'pre' condition
% Vhi_pre: High frequency signal from 'pre' condition
% Vlo_post: Low frequency signal from 'post' condition
% Vhi_post: High frequency signal from 'post' condition
% pval: 'theoretical' gives analytic p-values for R
% 'empirical' gives bootstrapped p-values for R
% ci: 'ci' gives confidence intervals, 'none' gives no confidence intervals
% varargin: optionally, include the parameter q indicating which quantiles
% of AmpLo you'd like to fit over
%
% OUTPUTS:
% XX.rpac: R_PAC value, confidence intervals XX.rPAC_CI
% XX.raac: R_AAC value, confidence intervals XX.rAAC_CI
% XX.rcfc: R_CFC value, confidence intervals XX.rCFC_CI
% XX.null: 3D surface for null model in Phi_low, A_low, A_high space
% XX.PAC: 3D surface for PAC model in Phi_low, A_low, A_high space
% XX.AAC: 3D surface for AAC model in Phi_low, A_low, A_high space
% XX.CFC: 3D surface for CFC model in Phi_low, A_low, A_high space
% P.rpac: p-value for RPAC statistic
% P.raac: p-value for RAAC statistic
% P.rcfc: p-value for RCFC statistic
% I.pval: p-value for indicator variable: low value implies significant effect of
% indicator on PAC
% I.beta: output beta values from full model including indicator
% variable
nCtlPts = 10;
%Compute phase and amplitude.
phi = [angle(hilbert(Vlo_pre));angle(hilbert(Vlo_post))];
amp = [abs(hilbert(Vhi_pre));abs(hilbert(Vhi_post))];
ampLO = [abs(hilbert(Vlo_pre)); abs(hilbert(Vlo_post))];
POST = [zeros(size(Vlo_pre)); ones(size(Vlo_post))];
%Define variables for GLM procedure.
Y = amp; %high frequency amplitude
X1 = spline_phase0(phi',nCtlPts); %low frequency phase
X2 = [ones(size(Y)),ampLO]; %low frequency amplitude
X3 = [X1,ampLO,sin(phi).*ampLO,cos(phi).*ampLO]; %low frequency phase, low frequency amplitude, interaction terms
X4 = [X3,X1.*POST];
XC = ones(size(Y)); %ones (null)
%Perform GLM.
[b1, dev1, stats1] = glmfit(X1, Y, 'gamma', 'link', 'log', 'constant', 'off'); %PAC
[b2, dev2, stats2] = glmfit(X2, Y, 'gamma','link','log','constant','off'); %AAC
[b3, dev3, stats3] = glmfit(X3, Y, 'gamma','link','log','constant','off'); %CFC
[bC, dev0, statsC] = glmfit(XC, Y, 'gamma', 'link', 'log', 'constant', 'off'); %null
[b4, dev4, stats4] = glmfit(X4, Y, 'gamma', 'link', 'log', 'constant', 'off');
%Chi^2 test between nested models (theoretical p-values)
chi0 = 1-chi2cdf(dev0-dev3,12);
chi1 = 1-chi2cdf(dev1-dev3,3); %Between PAC and PACAAC Model, if low AAC is present
chi2 = 1-chi2cdf(dev2-dev3,11); %Between AAC and PACAAC Model, if low PAC is present
chi3 = 1-chi2cdf(dev3-dev4,10);
I.pval = chi3;
I.beta = b4;
%create 3d model surfaces
phi0 = linspace(-pi,pi,100);
ampSORT = sort(ampLO);
if ~isempty(varargin)
q = varargin{1};
ind = find(ampLO>quantile(ampLO,q) & ampLO<quantile(ampLO,1-q));
ampSORT = sort(ampLO(ind));
end
XXC = []; XX1 = []; XX2 = []; XX3 = [];
ampAXIS = [];
YC = ones(size(phi0)); %null model
Y1 = spline_phase0(phi0',nCtlPts); %PAC model, function of phiLo
[splineC, ~, ~] = glmval(bC,YC,'log',statsC,'constant', 'off'); %null
[spline1, ~, ~] = glmval(b1,Y1,'log',stats1,'constant', 'off'); %PAC
count = 1;
for i = 1:100:length(ampSORT) %fit model over values of Alow
Y3 = [Y1,ampSORT(i)*ones(size(phi0))',ampSORT(i)*sin(phi0'),ampSORT(i)*cos(phi0')]; %CFC model, function of phiLo, ampLo, phiLo*ampLo
[spline3, ~, ~] = glmval(b3,Y3,'log',stats3,'constant', 'off');
XX3(:,count) = spline3;
ampAXIS(count) = ampSORT(i);
count = count+1;
end
L = length(1:100:length(ampSORT));
XX1 = repmat(spline1,1,L); %PAC model constant in PhiLow dimension
XXC = repmat(splineC,1,L); %null model constant in PhiLow, Alow dimensions
temp = ampSORT(1:100:end);
Y2 = [ones(size(temp)),temp];
[spline2,~,~] = glmval(b2,Y2,'log',stats2,'constant','off'); %AAC model, function of Alow
Xtemp = repmat(spline2,1,100); %AAC model constant in PhiLow dimension
XX.AAC = Xtemp'; XX.null = XXC; XX.PAC = XX1;XX.CFC = XX3; %3D model surfaces
XX.ampAXIS = ampAXIS; XX.phi0 = phi0; %axes
XX.rpac = max(max((abs(1-XX1./XXC))));
XX.raac = max(max((abs(1-Xtemp'./XXC))));
XX.rcfc = max(max((abs(1-XX3./XXC))));
if exist('pval','var') && strcmp(pval, 'empirical')
M = minvals(Vlo,Vhi); %find empirical p-values
P.rpac = max(.5,length(find(M.rpac>XX.rpac)))/length(M.rpac);
P.raac = max(.5,length(find(M.raac>XX.raac)))/length(M.raac);
P.rcfc = max(.5,length(find(M.rcfc>XX.rcfc)))/length(M.rcfc);
elseif exist('pval','var') && strcmp(pval, 'theoretical')
P.rpac = chi2; %use theoretical p-values
P.raac = chi1;
P.rcfc = chi0;
else
P = 'No p-values output';
end
if exist('ci','var') && strcmp(ci, 'ci')
phi0 = linspace(-pi,pi,100);
X0 = spline_phase0(phi0',nCtlPts);
Amax = max(ampSORT); Amin = min(ampSORT); stepsize = (Amax-Amin)/99;
X2eval = Amin:stepsize:Amax; %evaluate on all amplitudes
X2eval = [ones(size(phi0))',X2eval'];
%Determine CI for the measure RPAC.
M = 10000;
bMC = b1*ones(1,M) + sqrtm(stats1.covb)*normrnd(0,1,nCtlPts,M);
splineMC = glmval(bMC,X0,'log',stats1,'constant', 'off');
mx = zeros(M,1);
for k=1:M
mx(k) = max(abs(1-splineMC(:,k)./splineC));
end
r_CI = quantile(mx, [0.025, 0.975]);
XX.rpac_ci = r_CI;
%and for rAAC
M = 10000;
bMC = b2*ones(1,M) + sqrtm(stats2.covb)*normrnd(0,1,2,M);
splineMC = glmval(bMC,X2eval,'log',stats2,'constant', 'off');
mx = zeros(M,1);
for k=1:M
mx(k) = max(abs(1-splineMC(:,k)./splineC));
end
r2_CI = quantile(mx, [0.025, 0.975]);
XX.raac_ci = r2_CI;
%and for rCFC
M = 10000;
[m,~] = max(abs(1-XX3./XXC)); %find point of maximum distance between null, CFC models
[~,j] = max(m); %j ampLO, I(j) phiLO
bMC = b3*ones(1,M) + sqrtm(stats3.covb)*normrnd(0,1,nCtlPts+3,M);
Y1 = spline_phase0(phi0',nCtlPts); %model 1, function of phiLo
Y2 = [Y1,ampAXIS(j)*ones(size(phi0))']; %model 2, function of phiLo, ampLo
Y3 = [Y2,ampAXIS(j)*sin(phi0'),ampAXIS(j)*cos(phi0')];
splineMC = glmval(bMC,Y3,'log',stats3,'constant', 'off');
mx = zeros(M,1);
for k=1:M
mx(k) = max(abs(1-splineMC(:,k)./splineC));
end
r3_CI = quantile(mx, [0.025, 0.975]);
XX.rcfc_ci = r3_CI;
end
end
% Bootstrapped p-values
function M = minvals(Vlo,Vhi)
K = 100;
RPAC = zeros(1,K); RCFC = zeros(1,K); RAAC = zeros(1,K);
N = zeros(1,K); L = zeros(1,K);
for i = 1:K
Vhi_prime = AAFT(Vhi,1);
[XX] = glmfun(Vlo,Vhi_prime','none'); %compute R statistics between Vhi and shifted Vlo
RPAC(i) = XX.rpac;
RCFC(i) = XX.rcfc;
RAAC(i) = XX.raac;
end
M.rpac = RPAC; M.rcfc = RCFC; M.raac = RAAC; M.shiftN = N; M.shiftL = L;
end
% Generate a design matrix X (n by nCtlPts) for a phase signal (n by 1)
function X = spline_phase0(phase,nCtlPts)
% Define Control Point Locations
c_pt_times_all = linspace(0,2*pi,nCtlPts+1);
s = 0.5; % Define Tension Parameter
% Construct spline regressors
X = zeros(length(phase),nCtlPts);
for i=1:length(phase)
nearest_c_pt_index = max(find(c_pt_times_all<=mod(phase(i),2*pi)));
nearest_c_pt_time = c_pt_times_all(nearest_c_pt_index);
next_c_pt_time = c_pt_times_all(nearest_c_pt_index+1);
u = (mod(phase(i),2*pi)-nearest_c_pt_time)/(next_c_pt_time-nearest_c_pt_time);
p=[u^3 u^2 u 1]*[-s 2-s s-2 s;2*s s-3 3-2*s -s;-s 0 s 0;0 1 0 0];
X(i,mod(nearest_c_pt_index-2:nearest_c_pt_index+1,nCtlPts)+1) = p;
end
end