Identifying sequences in a Matrix

Currently working on a sequential ligament cutting sequence where we are trying to compare outcomes between similar clusters of sequential ligament cuts. For example we have

• Experiment 1: Lig A Lig B Lig C Lig D Lig E
• Experiment 2: Lig A Lig B Lig E Lig B Lig C
• Experiment 3: Lig A Lig E Lig B Lig C Lig D

With a separate matrix that are the measured outcomes corresponding to the sequence described

In comparing like clusters, we would like to compare outcomes that are similar i.e.

• Compare outcome after ligament A sectioning for all three groups (timepoint 1)
• Compare outcome after ligament A and B sectioning for Experiments 1 and 2 (timepoint 2)
• Compare outcome after ligament A/B/E sectioning with A/E/B in Experiments 2 and 3 (timepoint 3, order does not matter)

So far I've seen the diff function being used to determine whether, for example, ligament A and B are clustered, or ligament B and C are clustered, but this runs into problems when comparisons are made between A and C vs. C and E (both diff will produce 2). So this is clearly not a solution.

Any thoughts on how to tackle this? Thanks!

Answer 1

[Cedric]

You can find sequences using STRFIND, even when you are dealing with numbers:

 >> seq_str = 'ACEBAECE' ;
 >> strfind( seq_str, 'CE' )
 ans =
     2     7

 >> strfind( seq_num, [3,5] )
 ans =
     2     7

[RG]

One issue though - we are trying to do comparisons in which order does not matter. For instance if

Experiment A: Lig A Lig B Lig C Experiment B: Lig A Lig C Lig B

should be considered the same thing when using strfind, but this isn't the case. It would be easier if this was just three items, but with five time points it would be messy to write 120 OR statements.

[Cedric]

Use PERMS, or better, pre-process the sequence and replace all characters to match with some place holder, and search for place holders:

 seq_str  = 'ACEBAECE' ;          % Sequence.
 subseq   = 'CE' ;                % What we want to match but order doesn't matter.
 seq_copy = seq_str ;             % Build copy that we can alter.

Then we replace all letters from the copy that are in subseq with an underscore. Here is one way to do it if you have a recent enough version of MATLAB (let me know if not):

 seq_copy(any( seq_str == subseq.' )) = '_' ;

Evaluating the inner expressions helps to understand. There in an automatic expansion involved:

 >> seq_str == subseq.'
 ans =
  2×8 logical array
   0   1   0   0   0   0   1   0     <-- flag "has 'C'"
   0   0   1   0   0   1   0   1     <-- flag "has 'E'"

 >> any( seq_str == subseq.' )
 ans =
  1×8 logical array
   0   1   1   0   0   1   1   1     <-- flag "has any"

The outcome of the replacement is therefore this:

 seq_copy =
    'A__BA___'

And then we look up for occurrences of as many place holder characters as there are characters in subseq. There are two ways to do it depending if the sequence has to be "eaten" (which means that matched characters cannot be reused for matching what follows) or not:

 >> pos = regexp( seq_copy, repelem( '_', numel( subseq )))
 pos =
     2     6

 >> pos = strfind( seq_copy, repelem( '_', numel( subseq )))
 pos =
     2     6     7

and you have to pick the one that is appropriate for your context.