% Example code to extract and plot data from source data files for the
% paper "Receptive field center-surround interactions mediate context-dependent
% spatial contrast encoding in the retina" by Max Turner, Greg Schwartz,
% and Fred Rieke. eLife.
% 
% Prepared by Max Turner
% mhturner@stanford.edu
% 2018.09.03

%% Figure 2 - Source Data 1
% LinearEquivalentDiscModSurround is a structure array, each entry
% corresponds to one Off parasol RGC. For each cell, there are fields:
%   .Responses (no. images tested x no. surrounds tested). Each response
%       structure has matrices of individual trial spike binaries for image 
%       and linear equivalent disc stimuli
%   .Image (for each image tested)
%   .SurroundContrast (no. images tested x no. surrounds tested)
% The stimulus comes on at 200 msec and stays on for 200 msec

load('Figure 2 - Source Data 1.mat')
CellIndex = 1;
ImageIndex = 1;

%Filter parameters for converting spike binary trains to PSTH
sampleRate = 1e4;
PSTHsigma = 10; %msec
filterSigma = (PSTHsigma / 1e3) * sampleRate; %msec -> datapoints
newFilt = gaussFilter1D(filterSigma);
     
StimImage = LinearEquivalentDiscModSurround(CellIndex).Image{ImageIndex};
figure; clf;
for ss = 1:9
    subplot(1,10,ss); hold on;
    image_psth = sampleRate*conv(mean(LinearEquivalentDiscModSurround(CellIndex).Responses{ImageIndex,ss}.imageResponse),newFilt.amp,'same');
    disc_psth = sampleRate*conv(mean(LinearEquivalentDiscModSurround(CellIndex).Responses{ImageIndex,ss}.discResponse),newFilt.amp,'same');
    plot(image_psth,'k')
    plot(disc_psth,'r')
    axis off
    title(num2str(LinearEquivalentDiscModSurround(CellIndex).SurroundContrast(ss)))
end
subplot(1,10,10); imagesc(StimImage'); colormap(gray); axis image;


%% Figure 3 - Source Data 1, 2, 3
% Source Data 1: Fig. 3A-C
% Source Data 2: Fig. 3D-F
% Source Data 3: Fig. 3 - Suppl. 1

% FlashedGratingModulatedSurround is a struct array. Each entry corresponds
% to one cell. The fields of each structure are:
%   .GratingContrast = array of grating contrasts used for this cell
%   .SurroundContrast = array of surround contrasts used for this cell
%   .responses = (noGratingContrasts x noSurroundContrasts) cell array of
%       structs, each with a gratingResponse and a noGratingResponse for the
%       corresponding grating contrast and surround contrast. The extracellular
%       data contains both raw traces as well as binarized spike response
%       vectors
% The stimulus comes on at 200 msec and stays on for 200 msec

load('Figure 3 - Source Data 3.mat')
cellIndex = 3;
figure; clf; subplot(211); hold on;
plot(mean(FlashedGratingModulatedSurround(cellIndex).responses{4,5}.gratingResponse),'k')
plot(mean(FlashedGratingModulatedSurround(cellIndex).responses{4,5}.noGratingResponse),'r')
title('No surround')

subplot(212); hold on;
plot(mean(FlashedGratingModulatedSurround(cellIndex).responses{4,1}.gratingResponse),'k')
plot(mean(FlashedGratingModulatedSurround(cellIndex).responses{4,1}.noGratingResponse),'r')
title('+Surround')

%% Figure 4 - Source Data 1
% CenterSurroundWhiteNoise is a cell array, each entry of which is a structure that corresponds
% to a cell in the dataset. Structure fields:
% .stimulus = concatenated stimulus traces for .center and .surround
% stimuli
% .response = concatenated excitatory conductance response traces (in nS)
% for .center, .surround and .centerSurround stimuli
% Note that data are concatenated and grouped for convenience, but were
% acquired in interleaved trials
load('Figure 4 - Source Data 1.mat')

figure(10); clf;
for cc = 1:15
cellInd = cc;
sampleRate = 1e4; %Hz
filterLen = 500; %msec, length of linear filter to compute

%frequency after which to cut off filter spectrum. The stimulus updated at 30
%Hz, so the cutoff should be below that value to avoid ringing

filterPts = (filterLen/1000)*sampleRate; %msec -> datapoints

centerFilter = getLinearFilter(CenterSurroundWhiteNoise{cellInd}.stimulus.center,...
    CenterSurroundWhiteNoise{cellInd}.response.center);
centerFilter = centerFilter(1:filterPts); %trim to just filterLen

surroundFilter = getLinearFilter(CenterSurroundWhiteNoise{cellInd}.stimulus.surround,...
    CenterSurroundWhiteNoise{cellInd}.response.surround);
surroundFilter = surroundFilter(1:filterPts);  %trim to just filterLen

filterTimeVector = (1:filterPts) ./ sampleRate; %sec

subplot(4,4,cc); hold on;
plot(filterTimeVector, centerFilter,'b')
plot(filterTimeVector, surroundFilter,'r')
end

%% Figure 7 - Source Data 1
% CenterSurroundNaturalImageLuminance is a cell array, each entry is a
% structure corresponding to one cell. OFFcellInds and ONcellInds indicate
% the elements of the array that ar OFF and ON parasol RGCs, respectively.
% Each cell structure has stimuli and individual baseline-subtracted
% excitatory current responses for each of the stimulus conditions: center
% alone, surround alone, or center + surround in both control (correlated)
% and shuffled conditions
load('Figure 7 - Source Data 1.mat')
cellInd = 2;

figure; clf;
subplot(221); hold on;
plot(CenterSurroundNaturalImageLuminance{cellInd}.stimulus.controlCenter,'b')
plot(CenterSurroundNaturalImageLuminance{cellInd}.stimulus.controlSurround,'r')
title('Control')

subplot(222); hold on;
plot(CenterSurroundNaturalImageLuminance{cellInd}.stimulus.controlCenter,'b')
plot(CenterSurroundNaturalImageLuminance{cellInd}.stimulus.shuffleSurround,'r')
title('shuffled')

subplot(223); hold on;
plot(mean(CenterSurroundNaturalImageLuminance{cellInd}.response.controlCenterSurround),'k')
plot(mean(CenterSurroundNaturalImageLuminance{cellInd}.response.controlCenter) +...
    mean(CenterSurroundNaturalImageLuminance{cellInd}.response.controlSurround),'color',[0.7 0.7 0.7])

subplot(224); hold on;
plot(mean(CenterSurroundNaturalImageLuminance{cellInd}.response.shuffleCenterSurround),'k')
plot(mean(CenterSurroundNaturalImageLuminance{cellInd}.response.shuffleCenter +...
    CenterSurroundNaturalImageLuminance{cellInd}.response.shuffleSurround),'color',[0.7 0.7 0.7])

%% convenience functions...
function res = gaussFilter1D(sigma)
% Just a one-dimensional gaussian filter, for PSTH smoothing
x = -5*sigma:5*sigma;
amp = exp((-x.^2)./(2*sigma^2));
amp = amp./(sum(amp)); %integrates to 1

res.x = x;
res.amp = amp;
end


function LinearFilter = getLinearFilter(stimulus, response)
% Quickly estimate a linear filter using a white noise stimulus and response

FilterFft = mean((fft(response,[],2).*conj(fft(stimulus,[],2))),1) ;
LinearFilter = real(ifft(FilterFft)) ;

end