Question

I am trying to find an algorithm that does the following but have been unsuccessful:
I have a bunch of data that looks like the following:

Type    geneA    geneB    geneC    ...    geneN
A       1        0        1               1
A       0        0        1               1
B       1        1        1               0
C       0        1        1               0
B       1        1        0               0
etc        

So not all A’s are exactly the same, not all B’s are exactly the same, etc., but hopefully they have some kind of pattern. The data is preferably not just booleans (so would contain numbers other than just 0 and 1), but booleans would be fine for a start.

What I want to do is given a gene series for a person, find out what type they are.
For example I would like to input “011…1” and be told that this person is most likely type A.

This seems like something that should have been done before but I can’t seem to find any existing algorithms to do this (maybe pattern matching is the wrong key term?).

Any help on where to start looking for this kind of thing or algorithms that do this kind of thing would be much appreciated.

Answer 1

You could merge your gene* binary values into vectors: e.g. 1001...1, 001...1, etc. and perform k-means clustering on them.

For example, if you know a priori that you have three types A, B and C, you would likely perform clustering with k = 3.

Once you have clusters, you could use silhouettes to determine how well an input vector (e.g., 011...1) would fit to one of the three established clusters.