Suppose I have 5 traces of action potential with different peaks. I would like to plot all those 5 traces on the same plot so that all the peaks are same. Basically I would like to plot on the normalized scale.

I am trying to do this in matlab.

[hello]

My other question is that let's say I have 5 traces of action potentials again on the same graph. How can I graph a single plot that represents the average of these 5 traces of action potentials? In this case the time that it takes from start of action potential to reach the peak is different for different traces. That means time for one complete cycle of action potential might be different for the other cycle of action potential.

[dpb]

Well, then it matters what you mean by mean...hard to say without any data to even see, what more know what you would think would be the "right" answer given same.

I have no real idea what these "action potentials" represent; not a term am familiar with so what would be normal processing have no real idea. Examples, maybe??

[dpb]

Well, for anybody to do anything you'll have to attach the data.

And, be polite--don't keep asking same question over and over and over...

The figure is a single trace; you say you have five -- so is this just one of the five or are the individual peaks somehow to be separated/combined...

We only know what you tell us and can only work with what data is given.

[hello]

My apologies. I wrote it wrong. Actually if I count there are fifteen traces or cycles. I just want to get a single plot that is the average of all these fifteen traces.

[dpb]

Well, if you expect somebody to do something, you'll have to provide the data.

General idea would be to use findpeaks on negative of the trace to locate the minima and separate those out for processing. Again, just what you expect is extremely ambiguous at best.

[hello]

Atually my data is very big. I don't know if I can provide you the data but I separated into two traces as shown in the figure. My problem is that if i simply take the average it won't be correct because time that it takes for 1st trace to reach the peak is different than for the second trace to reach the peak. Simply taking an average might not even give me average peak value. Note that I manually picked the time interval.

[dpb]

How big is "very big"? You could certainly attach a .mat file with a third or half one would think without exceeding forum size limits.

Again, expecting anybody to be able to do anything useful just looking at figures is wishful thinking; it'll take somebody figuring out a way to find those breakpoints programmatically and then working out a scaling algorithm.

Again, what would you expect the results to look like if you were to do this by hand? Does it mean anything if align the peaks and what is it that is the actual result supposed to be? We still can only know what you tell us unless there is something about this in open literature that can point to...

Answer 1

[dpb]

If all are same length so have each trace in a column in an array, say data, then

 dataNorm=bsxfun(@rdivide, data, max(data));
 plot(x,dataNorm)

With data, a start:

[r,lag]=xcorr(Vv1,Vv2);    % cross-correlation, save lags
[~,ix]=max(abs(r));        % find maximum
lagDiff=lag(ix);           % and the lag number corresponding

figure                     % see what we started with...
subplot(2,1,1)
plot(Ti1.',[Vv1; Vv2].')   % both on same time scale
legend('Vv1','Vv2')
subplot(2,1,2)             % and after compensating for lag
plot(Vv1(lagDiff:end))     % the lagged Vv1
hold on
plot(Vv2)                  % compare to Vv2
legend('lagged Vv1','Vv2') % looks pretty good
delt=mean(Vv1(lagDiff:end)-Vv2(1:end-lagDiff+1));  % what's average offset
VV2=Vv2(1:end-lagDiff+1)+delt;   % and adjust Vv2 for that delta
plot(VV2,'cx',markersize',1.5)   % and add to see it, too...

Above yields--

Questions are just what is the purpose and what makes sense to do about normalizing magnitudes; what to actually consider if averaging makes sense and all that sorta' thingie...that has more to do with the subject matter under study and what the data mean than just the manipulations--they're pretty straightforward.

You will, it appears, need to do some smoothing to be able to locate peaks reliably in order to separate pulses; findpeaks had difficulty with the raw data as is in that regards.

[hello]

I am guessing you mean plot(x,dataNorm).

[dpb]

Ayup...originally overwrote data, then had change of mind but didn't follow thru. Good catch.

[hello]

Thank you for your help. I have a question why the numbers on the x axis are different for the second plot? So which one would be average plot of the given two traces?

[dpb]

Learn to read the code...what's the X value in the calls to plot between the two axes? How would you make the second look like the first and what would make more sense for a time axis when looking at the individual pulses than simple ordinal time?

As for the average, what would you propose averaging and why? What would an average mean?

[hello]

I am trying to get the average peak. I guess I could use findpeaks command. Let's say I have the average peak but what I would I do to the other y values. The problem is that I don't have equal time spacing for each pulses.

[dpb]

Well, we don't have a definition yet of what "average peak" means...the problem of duration is one of the issues, certainly.

What would it mean to average pulses that don't have similar durations and create some funky-looking shape? What purpose would that serve?

One could, of course, scale the various peaks to have a similar duration before averaging, but again, what would that actually then represent? Just because you can do something doesn't mean one should.

It just doesn't seem as though the problem description is well thought out as to what the end objective really is???

[dpb]

Again, for what definition of "amplitude"??? Have to define the problem before can solve it.

[hello]

When I gave you the first two pulses I made the time interval to be the same. But how could this be done if time interval are different? I should say my other goal is to also find the average time from start to the end of the average pulse.

[hello]

It would be peak amplitude.

[dpb]

"Peak" from what? FWHM, leading/trailing pulse edge, baseline of pulses, ... Many choices from which to choose.

Similarly to the amplitude scaling, one could scale (stretch/compress) the length between pulses to coincide but surely seems that would remove most if not all the actual physical meaning of the data to do so, wouldn't it? If you want everything to be identical, then it seems that just creating some arbitrary shape and using it irrespective of the nuances would be all that would be needed and there's really little point in actually acquiring data???

What am I missing here...

Again, though, if you do expect somebody to do something meaningful with one of these traces, will have to attach the trace for them to work with.

[hello]

Peak from the x axis. Yes you are right, we would not want to stretch/compress because we will lose meaning of the data. I think I have already attached the plot with the whole traces that I am trying to analyze.

[dpb]

"Peak from the x axis."

What does that mean????

The plot is of no use....only data will be of any benefit--and a coherent definition of what it is that you think you're really trying to find.

Answer 2

[dpb]

Well, another shot in the dark...putting the two traces you posted before together...

 VV=[Vv1 Vv2(2:end)]; T=[Ti1 Ti2(2:end)];  % concatenate data w/o overlap
 subplot(2,1,1), plot(T,VV)                % show it..
 xlim([34 37])
 subplot(2,1,2)                            % look at extents of sizable peaks
 findpeaks(VV,T,'minPeakProminence',20,'Annotate','extents')
 xlim([34 37]), ylim([0 40])

The results--

>> [pks,locs,widths,amps]=findpeaks(VV,T,'minPeakProminence',20,'Annotate','extents')
pks =
 37.1521   32.7119
locs =
 34.3990   35.8647
widths =
  0.4046    0.4049
amps =
 27.4835   27.4677
>> mean([amps.' widths.'])
ans =
 27.4756    0.4048
>>

You can try something of the sort on the whole trace and see what happens...

[hello]

Okay that is definitely helpful. Is there anyway I can find the time from the start of the pulse to the end of the pulse (I don't want to do this manually). Is the output "width" the max width of the pulse?

[dpb]

Read the documentation, look at the plot. The width and prominence are shown by the gold lines on the plot. The definition of prominence and width are given there in terms of how they're calculated; it's too involved to try to repeat here in brief.

As I said in first response, to try to find the actual pulses, I'd use findpeaks on the negative of the trace--

findpeaks(VV,T,'minpeakdistance',1)
[vlys,vlocs]=findpeaks(-VV,T,'minpeakdistance',1);
hold on
scatter(vlocs,abs(vlys),50,'r^','filled')
legend('Trace','Peaks','Valleys')

results in

How robust this will be in general, no klew...

ADDENDUM

Finding "the start of the pulse" again depends on what is the definition of start---the inflection point on the leading edge is going to be much more difficult undoubtedly, if that's it, because there's no real definitive transition; it's smooth transition albeit there is a very sharp positive gradient that might be able to localize...

[dpb]

NB: when look at the detail of the rise time and overlay the gradient the peak of the gradient, while looking like is dead-on in the wider view is, as expected, in the middle of the actual transition zone--it gets you in the neighborhood ok, but there's still a question of just what defines "the pulse", precisely.