ADSP PCA: Difference between revisions

From bradwiki
Jump to navigation Jump to search
No edit summary
Line 1: Line 1:
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.
PCA is a numerical transformation that performs dimensionality reduction, often for data visualization purposes. Like tSNE, PCA helps revealing clusters in high-dimensional data. PCA allows you to visualize two or three components directly against each at the same time.


{{SmallBox|float=right|clear=none|margin=0px 0px 8px 18px|width=170px|font-size=13px|Other Analyses|txt-size=11px|
{{SmallBox|float=right|clear=none|margin=0px 0px 8px 18px|width=170px|font-size=13px|Other Analyses|txt-size=11px|
Line 8: Line 8:
5. [[ADSP Stats|Descriptive Statistics]]<br>
5. [[ADSP Stats|Descriptive Statistics]]<br>
}}
}}
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: [https://github.com/subroutines/genos GENOS GIT]
Again, this code can be downloaded from the: [https://github.com/subroutines/genos GENOS GIT]

Revision as of 04:38, 6 February 2018

PCA is a numerical transformation that performs dimensionality reduction, often for data visualization purposes. Like tSNE, PCA helps revealing clusters in high-dimensional data. PCA allows you to visualize two or three components directly against each at the same time.

Other Analyses


Again, this code can be downloaded from the: GENOS GIT




PCA Code



% ######################################################################
%%                PRINCIPAL COMPONENTS ANALYSIS
% ######################################################################


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 = 50;


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  



%######################################################################
%%       PCA : PRINCIPAL COMPONENTS ANALYSIS
%######################################################################


ss = statset('pca')
ss.Display = 'iter';
ss.MaxIter = 100;
ss.TolFun = 1e4;
ss.TolX = 1e4;
ss.UseParallel = true;

% 'NumComponents',5,  'Algorithm','eig', Elapsed time is 70.459438 seconds.
% [PCAC,PCAS,PCAlatent,PCAtsquared,PCAE] = pca(  PCAMX' ,'Options',ss); 


% PCAMX(PCAMX>0) = 1;

[PCAC,PCAS,~,~,PCAE] = pca(  PCAMX' , 'Options',ss);


clearvars -except ADSP LOCI CASE CTRL PHEN PHE PHECASE PHECTRL...
AMX AMXCASE AMXCTRL ADNN PCAMX PCAC PCAS PCAE  







%% RECONSTRUCT ORIGINAL DATA (OPTIONAL TEST)



% MEAN DEVIATE (CENTER) DATA
mu = mean(PCAMX);
Xi = bsxfun(@minus,PCAMX,mu);


% Reconstruct the centered data
Xj = PCAS*PCAC';

% Reconstruct original data
PCAMX_REDUX = round(Xj' + mu);


clearvars -except ADSP LOCI CASE CTRL PHEN PHE PHECASE PHECTRL...
AMX AMXCASE AMXCTRL ADNN PCAMX PCAC PCAS PCAE PCAMX_REDUX




PCA Plots




ALL PCs in 1D

%% PLOT PCA --- ALL PCs IN 1D -------------------------------------
clc; close all;
fh1=figure('Units','normalized','OuterPosition',[.02 .06 .9 .8],'Color','w');
hax1 = axes('Position',[.08 .08 .40 .80],'Color','none');
hax2 = axes('Position',[.56 .08 .40 .80],'Color','none');


Tx = PHE.COHORTNUM;
Tn = unique(Tx);
Ts = 'Cohort';

% axes(hax1); colormap(hax1, (prism(max(Tn))) );
axes(hax1); colormap(hax1, flipud(jet(max(Tn))) );

ph1 = scatter( (1:numel(PCAC(:,1))).*0+1  , PCAC(:,1),50, Tx,'o');
hold on
ph2 = scatter( (1:numel(PCAC(:,2))).*0+2  , PCAC(:,2),50, Tx,'o');
hold on
ph3 = scatter( (1:numel(PCAC(:,3))).*0+3  , PCAC(:,3),900, Tx,'.');
hold on
ph4 = scatter( (1:numel(PCAC(:,4))).*0+4  , PCAC(:,4),900, Tx,'.');
hold on
ph5 = scatter( (1:numel(PCAC(:,5))).*0+5  , PCAC(:,5),900, Tx,'.');


hax1.XLim = [0 6];


text(1,max(ph1.YData),sprintf(['explains: \n ' num2str(round(PCAE(1),2)) '%% \n']),...
'Interpreter','tex','HorizontalAlignment','center','VerticalAlignment','bottom','FontSize',14);

text(2,max(ph2.YData),sprintf(['explains: \n ' num2str(round(PCAE(2),2)) '%% \n']),...
'Interpreter','tex','HorizontalAlignment','center','VerticalAlignment','bottom','FontSize',14);

text(3,max(ph3.YData),sprintf(['explains: \n ' num2str(round(PCAE(3),2)) '%% \n']),...
'Interpreter','tex','HorizontalAlignment','center','VerticalAlignment','bottom','FontSize',14);

text(4,max(ph4.YData),sprintf(['explains: \n ' num2str(round(PCAE(4),2)) '%% \n']),...
'Interpreter','tex','HorizontalAlignment','center','VerticalAlignment','bottom','FontSize',14);

text(5,max(ph5.YData),sprintf(['explains: \n ' num2str(round(PCAE(5),2)) '%% \n']),...
'Interpreter','tex','HorizontalAlignment','center','VerticalAlignment','bottom','FontSize',14);




xlabel('\bf PCA1 - PCA5');ylabel('\bf Eigenvector Coefficients');
cb=colorbar('Ticks',unique(Tx),'TickLabels',(unique(Tx)),'Direction','reverse');
cb.Label.String = Ts; cb.Label.FontSize = 18; cb.Label.Rotation = -90;
cb.Label.VerticalAlignment = 'baseline';





axes(hax2); colormap(hax2, (prism(max(Tn))) );

ph1 = scatter( (1:numel(PCAC(:,1))).*0+1  , PCAC(:,1),50, Tx,'o','MarkerEdgeAlpha',.02);
hold on
ph2 = scatter( (1:numel(PCAC(:,2))).*0+2  , PCAC(:,2),50, Tx,'o','MarkerEdgeAlpha',.02);
hold on
ph3 = scatter( (1:numel(PCAC(:,3))).*0+3  , PCAC(:,3),50, Tx,'o','MarkerEdgeAlpha',.02);
hold on
ph4 = scatter( (1:numel(PCAC(:,4))).*0+4  , PCAC(:,4),50, Tx,'o','MarkerEdgeAlpha',.02);
hold on
ph5 = scatter( (1:numel(PCAC(:,5))).*0+5  , PCAC(:,5),50, Tx,'o','MarkerEdgeAlpha',.02);

hax2.XLim = [0 6];

Top 50 variants Error creating thumbnail: File missing

Top 2000 variants Error creating thumbnail: File missing




ALZHEIMER'S STATUS

%% PLOT PCA --- CASE-vs-CTRL ------------------------------------------
clc; close all;
fh1=figure('Units','normalized','OuterPosition',[.02 .06 .97 .8],'Color','w');
hax1 = axes('Position',[.05 .07 .40 .85],'Color','none');
hax2 = axes('Position',[.52 .07 .40 .85],'Color','none');


Tx = PHE.AD;
Tn = unique(Tx);
Ts = 'CASE-vs-CTRL';


axes(hax1); colormap(hax1, (lines(max(Tn)+1)) );
ph1 = scatter(PCAC(:,1), PCAC(:,2),500, Tx,'.');
xlabel('\bf PCA1');ylabel('\bf PCA2');

cb=colorbar('Ticks',Tn,'TickLabels',Tn,'Direction','reverse');
cb.Label.String = Ts; cb.Label.FontSize = 14; cb.Label.Rotation = -90;
cb.Label.VerticalAlignment = 'baseline';


axes(hax2); colormap(hax2, (lines(max(Tn)+1)) );
ph2 = scatter3(PCAC(:,1), PCAC(:,2), PCAC(:,3),500, Tx,'.');
xlabel('\bf PCA1');ylabel('\bf PCA2');zlabel('\bf PCA3');
view([65,20])

Top 50 variants Error creating thumbnail: File missing

Top 2000 variants Error creating thumbnail: File missing




STUDY COHORT

%% PLOT PCA --- COHORT ------------------------------------------
clc; close all;
fh1=figure('Units','normalized','OuterPosition',[.02 .06 .97 .8],'Color','w');
hax1 = axes('Position',[.05 .07 .40 .85],'Color','none');
hax2 = axes('Position',[.52 .07 .40 .85],'Color','none');


Tx = PHE.COHORTNUM;
Tn = unique(Tx);
Ts = 'Cohort';


axes(hax1); colormap(hax1, flipud(jet(max(Tn))) );
ph1 = scatter(PCAC(:,1), PCAC(:,2),500, Tx,'.');
xlabel('\bf PCA1');ylabel('\bf PCA2');

cb=colorbar('Ticks',Tn,'TickLabels',Tn,'Direction','reverse');
cb.Label.String = Ts; cb.Label.FontSize = 14; cb.Label.Rotation = -90;
cb.Label.VerticalAlignment = 'baseline';


axes(hax2); colormap(hax2, flipud(jet(max(Tn))) );
ph2 = scatter3(PCAC(:,1), PCAC(:,2), PCAC(:,3),500, Tx,'.');
xlabel('\bf PCA1');ylabel('\bf PCA2');zlabel('\bf PCA3');
view([65,20])

Top 50 variants Error creating thumbnail: File missing

Top 2000 variants Error creating thumbnail: File missing

AGE

%% PLOT PCA --- AGE ------------------------------------------
clc; close all;
fh1=figure('Units','normalized','OuterPosition',[.02 .06 .97 .8],'Color','w');
hax1 = axes('Position',[.05 .07 .40 .85],'Color','none');
hax2 = axes('Position',[.52 .07 .40 .85],'Color','none');

AGE = round(PHE.AGE);
[Y,E] = discretize(AGE,8);
for nn = 1:numel(E)
AGE(Y==nn) = E(nn);
end

Tx = AGE(AGE>60);
Tn = unique(Tx);
Ts = 'AGE-STATUS';

PCAage = PCAC(AGE>60,:);


axes(hax1); colormap(hax1, (jet(numel(Tn))) );
% axes(hax1); colormap(hax1, flipud(jet(max(Tn))) );
ph1 = scatter(PCAage(:,1), PCAage(:,2),500, Tx,'.');
xlabel('\bf PCA1');ylabel('\bf PCA2');

cb=colorbar('Ticks',Tn,'TickLabels',Tn,'Direction','reverse');
cb.Label.String = Ts; cb.Label.FontSize = 14; cb.Label.Rotation = -90;
cb.Label.VerticalAlignment = 'baseline';


axes(hax2); colormap(hax2, (jet(numel(Tn))) );
% axes(hax2); colormap(hax2, flipud(jet(max(Tn))) );
ph2 = scatter3(PCAage(:,1), PCAage(:,2), PCAage(:,3),500, Tx,'.');
xlabel('\bf PCA1');ylabel('\bf PCA2');zlabel('\bf PCA3');
view([65,20])

Top 50 variants Error creating thumbnail: File missing

Top 2000 variants Error creating thumbnail: File missing



APOE STATUS

%% PLOT PCA --- APOE-STATUS ------------------------------------------
clc; close all;
fh1=figure('Units','normalized','OuterPosition',[.02 .06 .97 .8],'Color','w');
hax1 = axes('Position',[.05 .07 .40 .85],'Color','none');
hax2 = axes('Position',[.52 .07 .40 .85],'Color','none');


APOE = unique(PHE.APOE);
PHE.APOEID = PHE.SEX .* 0;
for nn = 1:numel(APOE)
    PHE.APOEID(PHE.APOE == APOE(nn)) = nn;
end


Tx = PHE.APOEID;
Tn = unique(Tx);
Ts = 'APOE-STATUS';


axes(hax1); colormap(hax1, (lines(numel(Tn))) );
% axes(hax1); colormap(hax1, flipud(jet(max(Tn))) );
ph1 = scatter(PCAC(:,1), PCAC(:,2),500, Tx,'.');
xlabel('\bf PCA1');ylabel('\bf PCA2');

cb=colorbar('Ticks',Tn,'TickLabels',APOE,'Direction','reverse');
cb.Label.String = Ts; cb.Label.FontSize = 14; cb.Label.Rotation = -90;
cb.Label.VerticalAlignment = 'baseline';


axes(hax2); colormap(hax2, (lines(numel(Tn))) );
% axes(hax2); colormap(hax2, flipud(jet(max(Tn))) );
ph2 = scatter3(PCAC(:,1), PCAC(:,2), PCAC(:,3),500, Tx,'.');
xlabel('\bf PCA1');ylabel('\bf PCA2');zlabel('\bf PCA3');
view([65,20])

Top 50 variants Error creating thumbnail: File missing

Top 2000 variants Error creating thumbnail: File missing
























Additional Genomics Analyses



Other Analyses










Notes

Category:ADSP

Retrieved from "https://bradleymonk.com/wiki/index.php?title=ADSP_PCA&oldid=3838"