ADSP PCA
t-Distributed Stochastic Neighbor Embedding (tSNE) is a technique like PCA that allows one perform dimensionality reduction for visualization purposes. Supposedly tSNE does better than PCA at revealing clusters in high-dimensional data. Whereas PCA only allows you to visualize two or three components directly against each at the same time -- tSNE uses math magic to coerce a high-dimensional dataset into either a 2D or 3D array.
t-SNE models each multi-dim object against a point on a euclidean surface in such a way that similar features are modeled by nearby point functions and dissimilar features are modeled by distant point functions. It then projects these points onto the plane allowing you visualize, what would effectively be, all the interesting principal component combinations - the ones that yield unique clusters - simultaneously.
Again, this code can be downloaded from the: GENOS GIT
t-SNE Code
% ######################################################################
%% tSNE : t-Distributed Stochastic Neighbor Embedding
% ######################################################################
clc; close all; clear; rng('shuffle')
cd(fileparts(which('GENOS.m')));
MATDATA = 'ADSPdata.mat';
which(MATDATA)
load(MATDATA)
clearvars -except ADSP
%% CARBON COPY MAIN VARIABLES FROM ADSP.STRUCT
LOCI = ADSP.LOCI(:,1:17);
CASE = ADSP.CASE;
CTRL = ADSP.CTRL;
PHEN = ADSP.PHEN;
clearvars -except ADSP LOCI CASE CTRL PHEN
%###############################################################
%% DETERMINE WHICH PARTICIPANTS TO KEEP
%###############################################################
PHE = PHEN(PHEN.TOTvars>14000,:);
PHECASE = PHE(PHE.AD==1,:);
PHECTRL = PHE(PHE.AD==0,:);
clearvars -except ADSP LOCI CASE CTRL PHEN PHE PHECASE PHECTRL
%###############################################################
%% COUNT NUMBER OF VARIANTS PER LOCI
%###############################################################
% The varsum() function will go through each known variant loci
% and check whether anyone's SRR ID from your subset of IDs match
% all known SRR IDs for that loci. It will then sum the total
% number of alleles (+1 for hetzy-alt, +2 for homzy-alt) for each
% loci and return the totals.
[CASEN, CTRLN] = varsum(CASE, PHECASE.SRR, CTRL, PHECTRL.SRR);
% SAVE COUNTS AS NEW TABLE COLUMNS
LOCI.CASEREFS = numel(PHECASE.SRR)*2-CASEN;
LOCI.CTRLREFS = numel(PHECTRL.SRR)*2-CTRLN;
LOCI.CASEALTS = CASEN;
LOCI.CTRLALTS = CTRLN;
clearvars -except ADSP LOCI CASE CTRL PHEN PHE PHECASE PHECTRL
%###############################################################
%% COMPUTE FISHER'S P-VALUE
%###############################################################
% COMPUTE FISHERS STATISTICS FOR THE TRAINING GROUP
[FISHP, FISHOR] = fishp_mex(LOCI.CASEREFS,LOCI.CASEALTS,...
LOCI.CTRLREFS,LOCI.CTRLALTS);
LOCI.FISHPS = FISHP;
LOCI.FISHORS = FISHOR;
clearvars -except ADSP LOCI CASE CTRL PHEN PHE PHECASE PHECTRL
%% MAKE LATEST COUNTS THE MAIN TABLE STATS
LOCI.CASEREF = LOCI.CASEREFS;
LOCI.CTRLREF = LOCI.CTRLREFS;
LOCI.CASEALT = LOCI.CASEALTS;
LOCI.CTRLALT = LOCI.CTRLALTS;
LOCI.FISHP = LOCI.FISHPS;
LOCI.FISHOR = LOCI.FISHORS;
%% SORT VARIANT LOCI TABLE BY FISHER P-VALUE
[X,i] = sort(LOCI.FISHP);
LOCI = LOCI(i,:);
CASE = CASE(i);
CTRL = CTRL(i);
LOCI.VID = (1:size(LOCI,1))';
LOCI.GENE = string(LOCI.GENE);
clc; clearvars -except ADSP LOCI CASE CTRL PHEN PHE PHECASE PHECTRL
disp(LOCI(1:9,:))
%% STORE VARIABLES FOR PCA/TSNE AS 'AMX'
AMX = LOCI;
AMXCASE = CASE;
AMXCTRL = CTRL;
clearvars -except ADSP LOCI CASE CTRL PHEN PHE PHECASE PHECTRL...
AMX AMXCASE AMXCTRL
%% FILTER VARIANTS BASED ALT > REF
PASS = (AMX.CASEREF > AMX.CASEALT./1.5) | (AMX.CTRLREF > AMX.CTRLALT./1.5);
sum(~PASS)
AMX = AMX(PASS,:);
AMXCASE = AMXCASE(PASS);
AMXCTRL = AMXCTRL(PASS);
AMX.VID = (1:size(AMX,1))';
clearvars -except ADSP LOCI CASE CTRL PHEN PHE PHECASE PHECTRL...
AMX AMXCASE AMXCTRL
%% TAKE THE TOP N NUMBER OF VARIANTS
N = 100;
AMX = AMX(1:N,:);
AMXCASE = AMXCASE(1:N);
AMXCTRL = AMXCTRL(1:N);
AMX.VID = (1:size(AMX,1))';
fprintf('\n %.0f final loci count \n\n',size(AMX,1))
clearvars -except ADSP LOCI CASE CTRL PHEN PHE PHECASE PHECTRL...
AMX AMXCASE AMXCTRL
%% MAKE RECTANGLE NN VARIANT MATRIX
[ADNN, caMX, coMX] = varmx(AMX,AMXCASE,AMXCTRL,PHE);
clearvars -except ADSP LOCI CASE CTRL PHEN PHE PHECASE PHECTRL...
AMX AMXCASE AMXCTRL ADNN
%% RANDOMIZE ADNN AND REORDER PHE TO MATCH ADNN
ADL = ADNN(1,:);
ADN = ADNN(2:end,:);
i = randperm(size(ADN,1));
ADN = ADN(i,:);
ADNN = [ADL;ADN];
[i,j] = ismember(PHE.SRR, ADN(:,1) );
PHE.USED = i;
PHE.ORDER = j;
PHE = PHE(PHE.USED,:);
PHE = sortrows(PHE,'ORDER');
PCAMX = ADNN(2:end,4:end);
clearvars -except ADSP LOCI CASE CTRL PHEN PHE PHECASE PHECTRL...
AMX AMXCASE AMXCTRL ADNN PCAMX
%% (OPTIONAL) PRE-PERFORM PCA BEFORE TSNE
% ss = statset('pca');
% ss.Display = 'iter';
% ss.MaxIter = 100;
% ss.TolFun = 1e4;
% ss.TolX = 1e4;
% ss.UseParallel = true;
%
% [PCAC,PCAS,~,~,~] = pca( PCAMX' , 'Options',ss);
% clc; close all; scatter(PCAC(:,1),PCAC(:,2))
%
% % ...,'NumPCAComponents',0,... means don't use PCA
% tSN = tsne(PCAC(:,1:10),'NumDimensions',2,'Theta',.6,'NumPCAComponents',0);
%
% clearvars -except ADSP GENB LOCI CASE CTRL PHEN AMX AMXCASE AMXCTRL...
% PHE ADNN PCAMX tSN PCAC PCAS
%######################################################################
%% tSNE : t-Distributed Stochastic Neighbor Embedding
%######################################################################
tSN = tsne(PCAMX,'NumDimensions',2,'Theta',.6,'NumPCAComponents',8);
disp('done')
clearvars -except ADSP GENB LOCI CASE CTRL PHEN AMX AMXCASE AMXCTRL...
PHE ADNN PCAMX tSN PCAC PCAS
t-SNE Plots
ALZHEIMER'S STATUS
%% PLOT TSNE --- ALZHEIMER'S STATUS (CASE/CTRL) --------------------------
close all;
fh1=figure('Units','normalized','Position',[.05 .05 .70 .84],'Color','w');
ax1=axes('Position',[.05 .02 .9 .9],'Color','none');
ph1 = gscatter(tSN(:,1),tSN(:,2), PHE.AD, [],'.',15);
title({'\fontsize{16} t-SNE : CASE vs CTRL',' '})
legend(ph1,{'CTRL','CASE'},'FontSize',12,'Box','off','Location','NorthWest');
axis off
Top 100 variants Error creating thumbnail: File missing
Top 2000 variants Error creating thumbnail: File missing
STUDY COHORT
%% PLOT TSNE --- CONSORTIUM STUDY COHORT (1:24) -------------------------
close all;
fh1=figure('Units','normalized','Position',[.05 .05 .70 .84],'Color','w');
ax1=axes('Position',[.05 .02 .9 .9],'Color','none');
ph1 = gscatter(tSN(:,1),tSN(:,2), PHE.COHORT, [],'.',15);
title({'\fontsize{16} t-SNE : STUDY COHORT',' '})
% legend(ph1,{'CTRL','CASE'},'FontSize',12,'Box','off','Location','NorthWest');
axis off
Top 100 variants Error creating thumbnail: File missing
Top 2000 variants Error creating thumbnail: File missing
SEX
%% PLOT TSNE --- SEX (M/F) ----------------------------------------------
close all;
fh1=figure('Units','normalized','Position',[.05 .05 .70 .84],'Color','w');
ax1=axes('Position',[.05 .02 .9 .9],'Color','none');
ph1 = gscatter(tSN(:,1),tSN(:,2), PHE.SEX, [],'.',15);
title({'\fontsize{16} t-SNE : SEX',' '})
legend(ph1,{'Male','Female'},'FontSize',12,'Box','off','Location','NorthWest');
axis off
Top 100 variants Error creating thumbnail: File missing
Top 2000 variants Error creating thumbnail: File missing
AGE
%% PLOT TSNE --- AGE (BINNED AGE) ---------------------------------------
close all;
fh1=figure('Units','normalized','Position',[.05 .05 .70 .84],'Color','w');
ax1=axes('Position',[.05 .02 .9 .9],'Color','none');
AGE = round(PHE.AGE);
ofAGE = AGE>60;
A = AGE(ofAGE);
histogram(AGE)
[Y,E] = discretize(A,[60 80 90 91]);
% [Y,E] = discretize(A,[60 75 85 90 91]);
for nn = 1:numel(E)
A(Y==nn) = E(nn);
end
ph1 = gscatter(tSN(ofAGE,1),tSN(ofAGE,2), A, [],'.',15);
title({'\fontsize{16} t-SNE : AGE',' '})
% legend(ph1,{'CTRL','CASE'},'FontSize',12,'Box','off','Location','NorthWest');
axis off
Top 100 variants Error creating thumbnail: File missing
Top 2000 variants Error creating thumbnail: File missing
APOE STATUS
%% PLOT TSNE --- APOE STATUS (22,23,24,33,34,44) ------------------------
close all;
fh1=figure('Units','normalized','Position',[.05 .05 .70 .84],'Color','w');
ax1=axes('Position',[.05 .02 .9 .9],'Color','none');
ph1 = gscatter(tSN(:,1),tSN(:,2), PHE.APOE, [],'.',15);
ph1(1).MarkerSize = 35;
ph1(2).MarkerSize = 25;
ph1(2).Color = [.20 .20 .99];
ph1(3).MarkerSize = 35;
ph1(4).Color = [.99 .50 .10];
ph1(5).Color = [.30 .70 .80];
ph1(6).MarkerSize = 25;
title({'\fontsize{16} t-SNE : APOE',' '})
% legend(ph1,{'CTRL','CASE'},'FontSize',12,'Box','off','Location','NorthWest');
axis off
Top 100 variants Error creating thumbnail: File missing
Top 2000 variants Error creating thumbnail: File missing
CONSENT GROUP
%% PLOT TSNE --- CONSENT GROUP ------------------------------------------
close all;
fh1=figure('Units','normalized','Position',[.05 .05 .70 .84],'Color','w');
ax1=axes('Position',[.05 .02 .9 .9],'Color','none');
ph1 = gscatter(tSN(:,1),tSN(:,2), PHE.RD, [],'.',15);
title({'\fontsize{16} t-SNE : CONSENT GROUP',' '})
% legend(ph1,{'CTRL','CASE'},'FontSize',12,'Box','off','Location','NorthWest');
axis off
Top 2000 variants Error creating thumbnail: File missing
Additional Genomics Analyses