Determine the subsets of data that are the most correlated with y-data

I have a pair of X-Y data containing 20 datapoints. Each datapoints correspond to a group, there are 8 groups, so, let say, X1 - X3 (together with Y1 - Y3) belongs to Group 1, and so on. The number of datapoints in each group is not necessarily the same. I found that for datapoints from, let say, Group 3 are very correlated between its X's and its Y's; but it's not good enough because I want to show that if I combine datapoints from multiple groups (let say Group 1, 3 and 7), I still can have high correlation. I dont think it's efficient to do this manually since I have to test 2^8=256 combinations. Is there any MATLAB function that do this automatically, or is there any math method to solve this automatically? Thank you.

[Adam Danz]

256 is a very manageable and relativly small number. Why not just compute all possible correlations in a loop (or loops)?