CAGE / RNA-seq normalizaton

Various normalization techniques in different tools

1. DESeq

Input a N by C matrix:
 N row ;  N genes
C column; Each column represents replica and different condition
Each cell : Represents the integer tag count value OR whatever value you want.

Algo:
1. From count matrix, for each row, calculate the geometric mean.
2. Diving each value by GM of corresponding row which generates size factor (SF) of each column.
3. Divide each count value by SF

MATLAB code

function doDESEQnorm

mymat=[   
    0    0    0    0    0    0    1;
    92    161    76    70    140    88    70;
    5    1    0    0    4    0    0;
    0    2    1    2    1    0    0 ];
   
% This shows four gene; in tow condition untreated and treated; First four replica for "untreated" and last %three column for "treated";


sizeRow = size(mymat,1);
sizeCol = size(mymat,2);

%% estimateSizeFactors %% 1 - Find Geometric Mean (GM) of Each row
gmRow = zeros(1,sizeRow);
for i=1:sizeRow
   
    curRow = mymat(i,:)';
    nzId = find(curRow);
    tmpVal = curRow(nzId);
    gm = geomean(tmpVal);
    gmRow(i) = gm;
   
end
mymat3= mymat;

%% estimateSizeFactors %% 2 - Divide each colum by corresponding row GM
afterDiv = mymat3;
for i=1:sizeRow
    afterDiv(i,:) = mymat3(i,:) / gmRow(i);
end



%% estimateSizeFactors %% 3 - SizeFactor: Take the median of Non-Zero Mormalized values
sizeFactor = zeros(1,sizeCol);
for col=1:sizeCol
   nzRow = find( afterDiv(:,col) );
   nzVal = afterDiv(nzRow,col);
   nzVal = sort(nzVal);
  
%    numberNZ = size(nzVal,1);
%    if rem(numberNZ, 2) ==0
%        idx = numberNZ/2 ;
%    else
%        idx = fix(numberNZ/ 2) +1;
%    end
%    sizeFactor(col)= nzVal(idx,1);

   sizeFactor(col) = median(nzVal);
  
end
sizeFactor

%% counts(sf, normalize=T) ; %% Do Normalzation
normValue = mymat;
for col=1:sizeCol
    normValue(:,col) = normValue(:,col)./sizeFactor(col);   
end

end