correlating many time series to find functional pairs

Hi there,

I'm a neuroscientist in deep water, trying to do things that aren't standard in my field. I have a strong feeling there are established methods for this, so please give some input if you can!

Data (very simplified but illustrates the issue):

- From 20 humans, I have derived 100 time series that reflect 100 brain processes.

- There should be the same 100 timeseries in all subjects, but I have no information on which ones are which. (With "the same" I mean they reflect the same underlying process and should therefore be highly temporally correlated despite the measurement noise.)


- I need a statistically valid way of "grouping" the timeseries across subjects, so that I end up with 100 timeseries that are averaged across 20 subjects and each reflect the same process.

Does this make sense?

I know I can do this in 2 subjects by calculating a cross correlation matrix, but I'm sure there are more elegant/valid ways of working with 20 subjects without doing many pairwise correlations... (and I wouldn't know how do deal with that output).

I've thought about seed-based correlation analysis, i.e. picking a seed timecourse in one subject and correlating it with everything, then changing the seed and repeating etc. But again I'm not sure where I'm going with that, and this application is very different from what seed based correlation is usually used for with fMRI data.

I barely know what to google for, so I'm hoping that someone here can throw some keywords at me, or come up with some ideas (wild ones too) about the best ways of doing this might be!

I use Matlab and am learning R. I don't mind learning complex methods, but need a place to start.

Thanks a lot!