Signal analysis

Discussion in 'Programmer's Corner' started by nico, May 24, 2007.

  1. nico

    Thread Starter New Member

    May 24, 2007
    4
    0
    Hi everyone!

    I'm not sure if this is the best place to ask this question, but I'll give it a try.

    So, I'm facing a quite complex problem, that I will try to summarize here, and I would like to hear any ideas on how I can possibly get around it.

    I have recordings of 2 different signals: in particular one is a recording of current passing through an electrode and the other is a recording of fluorescence intensity recorded with a CCD camera. These recordings can be as long as 1 or 2 hours and, for technical reasons, have to be recorded at different rates, in particular 1Hz for the fluorescence and 1kHz for the current.

    Now, in some occasions there are definite events in these signals, seen as transient increases in the fluorescence or in the current. I want to see if the times when these event happens are the same or not. Just by looking at the traces I would say that this is almost always the case, but I would like to have an automatic procedure to do that.

    Any idea of how to solve this problem? I was reading about correlation functions, would that be of any help to me? And do the different sampling rates represent a problem?

    Also, to make things more complicated... the two types of increases are different. While the fluorescence just increases, stays high for 5-10 seconds and then goes down, the electrical signal has only very quick spikes during that period (say, 50-100 spikes during those 5-10 seconds).

    Well, I hope all of this makes some sense to you guys, and that you can give me some ideas!

    Thank you a lot
    nico
     
  2. beenthere

    Retired Moderator

    Apr 20, 2004
    15,815
    282
    Those sorts or correlations can be tedious. You could devise a circuit to flash an LED and put in a distinctive additional signal to the electrical recording, but the differing recording rates are going to make that difficult. The signals would have to be of short duration in order to be good time markers. That makes it easy to lose the flash of the time marker in the fluorescence data.

    For that matter, you can lose a lot of flicker at a 1 Hz rate. I've made event markers for video and electrical activity, but the recording rates made it possible to be confident of the time correlations.

    Are you averaging the intensity over time? That will make it very difficult to say just when the intensity changed. I wonder is using a device like a TAOS light-to-voltage converter wouldn't make it easier to do the recordings at the same rate. That would make event marking or timing tics pretty simple to accomplish.
     
  3. nico

    Thread Starter New Member

    May 24, 2007
    4
    0
    Hi beenthere, thank you for your reply.

    First of all, I have to say, I'm a biologist, so I would probably be quite clueless on how to create a circuit as you said :p
    The 1 Hz recording is as far as we can get (we can go up to 2 Hz in some cases, but it's not so easy) because of the kinetics of the fluorophore that we're using. Anyway the changes are big and long enough for us to be able to distinguish them from background noise.
    I didn't mention that these signals are recorded from living cells, and they can be quite tricky to obtain, so usually we perform all the analysis offline.
    Our workflow is usually to save the recorded traces on the computer, put the values in a DB and use a custom written software to analyze them, after the experiment has finished.

    The fluorescence is recorded with an exposure time in the order of 200-300 ms per frame, with a CCD camera attached to a microscope.

    nico
     
  4. beenthere

    Retired Moderator

    Apr 20, 2004
    15,815
    282
    A brief web search indicates that the fluorphore emission is likely to be low-level, which may have to do with the long collection period (gotta have enough photons to work with). Too bad you're using a ccd camera - something like a photomultiplier tube would be more sensitive and allow faster recording, if significantly more difficult to use.

    Do I correctly infer that this is not simple stimulus-response? More like a voltage clamp setup to record the cell's change in electrical activity?

    Whatever the case, it will be difficult to unequivocally relate the two data streams. It might be possible to make a timing marker as I suggested earlier, but relating a video frame to the electrical activity could be problematic, unless that activity takes place at intervals far enough apart in time that the intensity variance can be demonstated to be related to the electricrode activity.

    Rather than flood the microscope's field with flashes, one refinement might be to run a small diameter optical fibre to a corner of the camera's field of view. That way, you can get a time marker in the visual field without washing out the fluorescence.
     
  5. nico

    Thread Starter New Member

    May 24, 2007
    4
    0
    You're right, I'm doing voltage clamp and (unfortunately) the fluorescence is quite dim (this can't be helped for various problems related more to the biology of the system than to physical matters).

    I forgot to say that the two recordings are absolutely syncronized, as the two softwares for the acquisition of the data run toghether: you press one button in software 1 and this start both the recording of fluorescence in software 1 and that of current in software 2.
    My question is therefore more on mathematical methods to say: if I have a function with peaks at 18, 20, 25 and 30 sec and another function with a train of peaks (ie.: 50-100 events) between 18.2-18.5, 20.1-20.6, 25.3-25.9, 30.5-31.0 sec are the various events related? It's obvious looking at the numbers, but I don't have a clue what's the best way to analyze that mathematically.

    Thank you very much for your suggestions
    nico

    PS: we have also another setup that uses a photomultiplier tube, but we actually find out it to be less sensitive than the other system, at least for what we want to do. :(
     
  6. beenthere

    Retired Moderator

    Apr 20, 2004
    15,815
    282
    I'd better bow out at this point. I'm mostly hardware oriented. I have written code to run interfaces and small operating systems, but no number crunching.

    Good luck!
     
  7. nico

    Thread Starter New Member

    May 24, 2007
    4
    0
    Ok, don't worry, thank you a lot anyway. Maybe someone else has some ideas?
     
Loading...