how to compare every pair of rows and return a matrix?

Hello guys, I have a matrix of 135 rows and 2 columns and I want to compare every two pairs of rows based on their overlapped elements and return a matrix.

 I have attached some rows from my matrix to make the question more clear

[KSSV]

An example with the data will help us to help you.....what you want to compare in the rows?

[chocho]

This is my data Bro you see in the example i posted there are 10 rows, I want to compare the elements of column1 based on column 2 elements.

 like row 1 with all the remaining rows then row2 with all the rest f the rows.

For example, row1 and row2 have no common elements, row 1 and row 3 have also no common elements.

However, if i take row2 and row2 they have 5 common elements.

So my expected output is 10*10 matrix, in which the elements or the cells of this matrix represent the number of the common elements.

I hope you get my idea

[Guillaume]

It's not clear what a common element is. Not helped by the fact that there's probably a typo in if i take row2 and row2 they have 5 common elements since comparing something with itself is certain to say that everything is common.

Similarly, defines what compare means for you.

[chocho]

@Guillaume sorry is a typo I mean row2 and row3 because they have 5 genes in common (RASGRP4, RASGPR2, RAF1, PRAKACA, RAS1) so the output is 5

Answer 1

[Jan]

I guess your matrix is a cell string and you want to find common sub-strings in it.

% Assuming your matrix is called "M":
n = size(M, 1);

% Convert the strings in the 2nd column to cell strings:
Sub = cell(1, n);
for k = 1:n
  Sub{k} = strtrim(strsplit(M{k, 2}, ','));
end

Match = cell(n, n);
for i1 = 1:n
  for i2 = i1+1:n
    v             = intersect(Sub{i1}, Sub{i2});
    Match{i1, i2} = v;
    Match{i2, i1} = v;   % Output is symmetric
  end
end

Or maybe you want ismember instead of intersect. Or you want to output the indices? Or the output should be a logical matrix, which is TRUE if the sub-strings have any overlap?

Please explain "compare every two pairs of rows" specifically. Most likely the above code can be adjusted easily.

[chocho]

@Jan Bro yes i want to output numbers. if 2 rows have 5 common genes so is 5

[chocho]

@Jan Bro see i have 2 columns right and all of 10 rows.

Column1 have 10 rows right and these rows have nothing to compare between them but have some common elements in column 2. So i want to compare the elements of column2 one by one like row 1 with row2.......row10 and then row2 with row1 row2 .......row10 and then row3 with row1 row2 .......row10 ........ and row10 with row1 row2 and .......row10 .

So the output is a matrix of size 10*10 their cells the number of the common genes between every pairs of rows. may be 10 may be 2 or 1 or even 0.

[Jan]

@chocho: I cannot follow your explanations. What does "i have 2 columns right and all of 10 rows" mean? I guess I understand at least, that you want only the number of intersections. Then:

Match = zeros(n, n);
for i1 = 1:n
  for i2 = i1+1:n
    v             = sum(ismember(Sub{i1}, Sub{i2}));
    Match(i1, i2) = v;
    Match(i2, i1) = v;   % Output is symmetric
  end
end

What do you want on the diagonal? Maybe insert

for i1 = 1:n
  Match(i1, i1) = numel(Sub{i1});
  ...

By the way, I do not like to be called "Bro".

[chocho]

@Jan So sorry for that never repeat it

yes, you got what i exactly needed

Last questions plz if i want to change v by (1/(1/(v/m))+(1/(v/n))) where v is the number of intersections and m is the number of strings m row and n is the number of strings of n row

For example, if we take row 1 of m=3 sub_strings and row2 of n=10 sub-strings and v is the number of intersections which already calculated above So "new_variable=(1/(1/(v/m))+(1/(v/n)))"

[chocho]

@Jan Any help for the last question?

[Jan]

What exactly is the question? You have posted working code already. With a simplification:

new_variable = v / (m + n);

Or as complete code (I use "len1" and "len2" instead of "n" and "m"):

Match = zeros(n, n);
for i1 = 1:n
  len1 = numel(Sub{i1});
  for i2 = i1+1:n
    len2          = numel(Sub{i2});
    v             = sum(ismember(Sub{i1}, Sub{i2})) / (len1 + len2);
    Match(i1, i2) = v;
    Match(i2, i1) = v;   % Output is symmetric
  end
end

[chocho]

@ Jan hello jan plz if I want to change new_variable = v / (m + n) to new_variable = v^2 / (m * n) So this line will be like this :

      v= sum(ismember(Sub{i1}, Sub{i2}))^2 / (len1 * len2);

is it right ??

[Jan]

Yes.