#############1: differential expression
clear;
data=csvread('gene_expression_254.csv');%'miRNA_expression_250.csv'
group=data(1,:);
exp=data(2:end,:);
[ngenes ns]=size(exp);
clear p p2 ;
for r=1:ngenes
    [h(r) p(r)]=ttest2(exp(r,group==1),exp(r,group==0));
    [p2(r), h2(r)] = ranksum(exp(r,group==1),exp(r,group==0));%runksum test is equivalent to a Mann-Whitney U-test
end

xlswrite('pvalue_gene.xlsx',cat(2,p',p2'),1);%'pvalue_miRNA.xlsx'


################2: linear regression
clear;
    
fidw=fopen('miRNA_gene_regression_247.csv','w');       % OUTput file name by part;
fmt_record='%s,%s,%e,%e,%e\n';

[miRData miRName]=xlsread('miRNA_cc_247_human.xlsx');
[nmiRs nsamples]=size(miRData);

% Input expression data and gene name;
[Y GeneName]=xlsread('gene_cc_247.xlsx');
nGenes=size(Y,1);


% Calculation and output regression results;
CVALUE=1E-4;
for i=1:nmiRs
    X=cat(2,ones(nsamples,1),miRData(i,:)');
    results=olsvectory(Y',X);
    for j=1:nGenes
        clear pv record;
        pv=results.pvalue(2,j);
        if pv<CVALUE
        record{1}=miRName{i};
        record{2}=GeneName{j};
        record{3}=results.beta(2,j);
        record{4}=results.pvalue(2,j);
        record{5}=results.rsqr(j);
        fprintf(fidw,fmt_record,record{1},record{2},record{3},record{4},record{5});   
        end
    end
end
fclose(fidw);
clear;


%%%%%%%% for missing data, replace with mean
function W=missing(X)

m=size(X,1);
for i=1:m
    x=X(i,:);
    mu=mean(x(~isnan(x)));
    x(isnan(x))=mu;
    X(i,:)=x;
end
W=X;  

%%%%%% normalize data by minus mean and divided by std;
% normalize data by row;
% reduce memory occupation;
function Y=normalization(E)
[m n]=size(E);
mu=mean(E,2);
sd=std(E,0,2);

Y=[];
    for j=1:n
        clear C;
        C=E(:,j)-mu;
        Y=cat(2,Y,C./sd);
    end


%%%%%% unnormal the data
function [W,I]=del_unnormal(X)

[m_X,n_X]=size(X);
Itmp=ones(1,m_X);
for i=1:m_X
    tmp=X(i,:);
    if length(tmp(isnan(tmp)))>=n_X/2 || std(tmp(~isnan(tmp)))<1e-10
        Itmp(i)=0;
    end
end
I=find(Itmp==1);
I=I';
W=X(I,:);


%%%%%%%%%% standardization for X according to rows
% X m*n matrix
% W m*n matrix
function W=standardization(X)

[m,n]=size(X);
mu=mean(X,2);
S=std(X,0,2);
X=X-repmat(mu,1,n);
W=diag(1./S)*X;

end



function results=olsvectory(y,x)
% PURPOSE: least-squares regression 
%---------------------------------------------------
%  ???? ?????y?x, ??????????
%
% USAGE: results = olsvectory(y,x)
% where: y = dependent variable vector    (nobs x m)
%        x = independent variables matrix (nobs x nvar)
%---------------------------------------------------
% RETURNS: a structure
%        results.meth  = 'ols'
%        results.beta  = bhat     (nvar x m)
%        results.tstat = t-stats  (nvar x m)
%        results.pvalue= p-values (nvar x m)
%        results.bstd  = std deviations for bhat (nvar x m)
%        results.yhat  = yhat     (nobs x m)
%        results.resid = residuals (nobs x m)
%        results.sige  = e'*e/(n-k)   (1*m) 
%        results.rsqr  = rsquared     (1*m)  
%        results.rbar  = rbar-squared  (1*m) 
%        results.dw    = Durbin-Watson Statistic
%        results.nobs  = nobs
%        results.nvar  = nvars
%        results.y     = y data vector (nobs x m)
%        results.bint  = (nvar x2 ) vector with 95% confidence intervals on beta
%Warning!!!!!!  R-square and the F statistic are not well-defined unless x has a column of ones.
%---------------------------------------------------
% SEE ALSO: prt(results), plt(results)
%---------------------------------------------------

% Change by Xiangzhong Fang from the program by

% James P. LeSage, Dept of Economics
% University of Toledo
% 2801 W. Bancroft St,
% Toledo, OH 43606
% jlesage@spatial-econometrics.com
%
% Barry Dillon (CICG Equity)
% added the 95% confidence intervals on bhat

if (nargin ~= 2); error('Wrong # of arguments to ols'); 
else
 [nobs nvar] = size(x); 
 [nobs2 m] = size(y);
 if (nobs ~= nobs2); error('x and y must have same # obs in ols'); 
 end;
end;

results.meth = 'ols';
results.y = y;
results.nobs = nobs;
results.nvar = nvar;

if nobs < 10000
  [q r] = qr(x,0);
  xpxi = (r'*r)\eye(nvar);
else % use Cholesky for very large problems
  xpxi = (x'*x)\eye(nvar);
end;

results.beta = xpxi*(x'*y);
results.yhat = x*results.beta;
results.resid = y - results.yhat;
sigu = sum(results.resid.*results.resid);%m*1 vector
results.sige = sigu/(nobs-nvar);
tmp = (diag(xpxi))*(results.sige); 
sigb=sqrt(tmp);
results.bstd = sigb;
%tcrit=-tinv(1-.025,nobs);
%results.bint=[results.beta-tcrit.*sigb, results.beta+tcrit.*sigb];
results.tstat = results.beta./(sqrt(tmp));
ym = y - repmat(mean(y),nobs,1);
rsqr1 = sigu;
rsqr2 = sum(ym.*ym); %m*1 vector
results.rsqr = 1.0 - rsqr1./rsqr2; % r-squared
%rsqr1 = rsqr1/(nobs-nvar);
%rsqr2 = rsqr2/(nobs-1.0);
%if rsqr2 ~= 0
%results.rbar = 1 - (rsqr1/rsqr2); % rbar-squared
%else
%    results.rbar = results.rsqr;
%end;

results.pvalue=2*(1-tcdf(abs(results.tstat),nobs-nvar));%Fang

%ediff = results.resid(2:nobs) - results.resid(1:nobs-1);
%results.dw = (ediff'*ediff)/sigu; % durbin-watson