[Eeglablist] std_envtopo (plotchans and pvaf)

[Ana Navarro Cebrian]

Hi Makoto,

Maybe you want to try my toolbox to backproject cluster ERPs to channels.
> I'll send it to you in the separate mail.
>
I'd love to. Thank you.

I saw it. It is quite strange indeed. I want to diagnose the situation.
> At least I can tell that your 'outermost envelope' (shown in thick black)
> is surpassed by other envelopes, which is not normal. That is like saying
> some of your clusters have larger values than the sum of all (it's not
> exactly like this but pretty much about it). Can you think of any reason
> why you see this?
>
> If you can't find anything wrong by yourself, I need to ask you to either
> 1. transfer all of your data to our server or 2. video conference by
> sharing your desktop. You can choose whichever you want. Sorry for
> inconvenience.
>
That is strange. I have no idea what could be wrong. I tried clustering
again with a greater (20) and a smaller (2) number of clusters. I get the
same two clusters both ways and the outermost envelope is always surpassed
by the envelopes of these two clusters, even when those two clusters are
the only ones in the analysis. The only thing I can think of is that these
two clusters are of an opposite signal and they kind of cancel each other
out, so they are greater alone than together. This would be kind of related
to the problem of having all the electrodes included. I'm not sure if that
makes sense.
I'm happy to transfer my data so you can take a look at it. I just looked
at only 3 subjects and I get the same effect. I could transfer those
instead of the entire group if you want.
Let me know.
Thanks,
Ana


>
> Makoto
>
>
> 2013/4/10 Ana Navarro Cebrian <anavarrocebrian at gmail.com>
>
>> >Currently this is not supported. I see how to do it though.
>> I would suggest that you exclude clusters of non-interest, or specify the
>> clusters to use, so that your outermost envelope is consist of only
>> necessary ones.
>>
>> Thanks Makoto. This is very helpful. I'd still love to be able to choose
>> some channels though; My P300 comes along with an effect of an opposite
>> (negative) signal in the frontal electrodes (I think this happens because I
>> use the average as a reference), and I believe that choosing just some
>> centro-parietal electrodes would give me a better estimate. I'll see if I
>> can try something myself.
>>
>> >-550.46 should be a number of latency. Where does the line (extending
>> from the scalp topos) pointing? Isn't it -550.46 ms? If so, you should
>> limit the latency window to evaluate contribution.
>>
>> It seems to me that the line is pointing at 559 and what I'm talking
>> about says "pvaf: 550.46. I had limited the latency anyway (from 200 to
>> 600ms). Another example is the cluster close to it, which shows "pvaf:
>> -111.61". This one is also pointing at latency ~559ms. I've attached an
>> example of the output I get from std_envtopo.
>>
>> >No, actually pvaf never sums to 100% if you add up each clusters. That
>> means, pvaf(Cls1+Cls2) ~= pvaf(Cls1)+pvaf(Cls2). By the way the default
>> 100% is pvaf(Cls1+Cls2+...ClsN) if you have N number of clusters.
>> The measure pvaf is always superadditive i.e. exceeds 100% if summed
>> separately.
>>
>> Here I was talking about the individual pvaf (for each individual
>> cluster) shown under the cluster topoplot, and not the total pvaf. I'm not
>> sure if the response is still the same.
>> For example, for those two clusters that I'm talking about above, with
>> pvaf -550.46% and -111.61%. Would it be possible to get positive values
>> like those for individual clusters? i.e. pvaf: 550.46?
>>
>> Thank you again for your help, Makoto.
>>
>>
>>
>>
>> 2013/4/10 Makoto Miyakoshi <mmiyakoshi at ucsd.edu>
>>
>>> Dear Ana,
>>>
>>> >First, I was wondering wether I could use something like the
>>> 'plotchans' (available in envtopo) to compute the contributions in just one
>>> or a few channels of interest. The default function computes the
>>> contributions for the grand ERPs which gives me a lot of variability that
>>> I'm not interested in.
>>>
>>> Currently this is not supported. I see how to do it though.
>>> I would suggest that you exclude clusters of non-interest, or specify
>>> the clusters to use, so that your outermost envelope is consist of only
>>> necessary ones.
>>>
>>> >Also, I'm still confuse about the pvaf for the individual clusters.
>>> For example, I'm looking for the clusters that explain the P300
>>> variability and there is a cluster (out of 7 clusters) that seems to me
>>> (based on the cluster's ERP) that explains most of the variability of the
>>> P300.
>>> When I run std_envtopo, I get a pvaf of -550.46 for that cluster.
>>> First, I understand that this is not talking about the P300 activity alone,
>>> but the grand average, and this implicates a lot of variance from many
>>> areas that I'm not interested in (and this is the reason why I'm trying to
>>> use just a few channels of interest). Therefore, because what I think is
>>> the 'P300 cluster' may have a different signal than all the other
>>> activities (that I'm not interested in), then I get a negative value for
>>> this cluster pvaf (-550.46). Am I getting something wrong so far?
>>>
>>> -550.46 should be a number of latency. Where does the line (extending
>>> from the scalp topos) pointing? Isn't it -550.46 ms? If so, you should
>>> limit the latency window to evaluate contribution.
>>>
>>> >Also, I imagine that, because this is the percent variance accounted
>>> for, for positive numbers, 100 should be the maximum possible value?
>>>
>>> No, actually pvaf never sums to 100% if you add up each clusters. That
>>> means, pvaf(Cls1+Cls2) ~= pvaf(Cls1)+pvaf(Cls2). By the way the default
>>> 100% is pvaf(Cls1+Cls2+...ClsN) if you have N number of clusters.
>>>
>>> The measure pvaf is always superadditive i.e. exceeds 100% if summed
>>> separately.
>>>
>>> Makoto
>>>
>>>  2013/4/10 Ana Navarro Cebrian <anavarrocebrian at gmail.com>
>>>
>>>> Hi,
>>>>  I have two questions about the function std_envtopo.m
>>>>
>>>> First, I was wondering wether I could use something like the
>>>> 'plotchans' (available in envtopo) to compute the contributions in just one
>>>> or a few channels of interest. The default function computes the
>>>> contributions for the grand ERPs which gives me a lot of variability that
>>>> I'm not interested in.
>>>>
>>>> Also, I'm still confuse about the pvaf for the individual clusters.
>>>> For example, I'm looking for the clusters that explain the P300
>>>> variability and there is a cluster (out of 7 clusters) that seems to me
>>>> (based on the cluster's ERP) that explains most of the variability of the
>>>> P300.
>>>> When I run std_envtopo, I get a pvaf of -550.46 for that cluster.
>>>> First, I understand that this is not talking about the P300 activity alone,
>>>> but the grand average, and this implicates a lot of variance from many
>>>> areas that I'm not interested in (and this is the reason why I'm trying to
>>>> use just a few channels of interest). Therefore, because what I think is
>>>> the 'P300 cluster' may have a different signal than all the other
>>>> activities (that I'm not interested in), then I get a negative value for
>>>> this cluster pvaf (-550.46). Am I getting something wrong so far?
>>>>
>>>> Also, I imagine that, because this is the percent variance accounted
>>>> for, for positive numbers, 100 should be the maximum possible value?
>>>>
>>>> I hope that makes sense. Thanks in advance for your help.
>>>>  Ana
>>>>
>>>> _______________________________________________
>>>> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
>>>> To unsubscribe, send an empty email to
>>>> eeglablist-unsubscribe at sccn.ucsd.edu
>>>> For digest mode, send an email with the subject "set digest mime" to
>>>> eeglablist-request at sccn.ucsd.edu
>>>>
>>>
>>>
>>>
>>> --
>>> Makoto Miyakoshi
>>> Swartz Center for Computational Neuroscience
>>> Institute for Neural Computation, University of California San Diego
>>>
>>
>>
>>
>> --
>> Ana Navarro-Cebrian
>> Postdoctoral Fellow. UCSF
>>
>
>
>
> --
> Makoto Miyakoshi
> Swartz Center for Computational Neuroscience
> Institute for Neural Computation, University of California San Diego
>



-- 
Ana Navarro-Cebrian
Postdoctoral Fellow. UCSF
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20130418/36c8361f/attachment.html>