 |
 |
Hi Rita, Sorry for the delay in replying. Let me summarise my understanding of your question. You’d like your driving input region to have no difference between conditions, so it’s a simple main effect of task. You’d then like the effect of stimulus to modulate the connection to your second region, thereby forming an interaction with the driving input. You’re concerned because both regions are modulated by condition, which doesn’t match the attention example. Let’s forget the attention example for now and concentrate on your data. There are different possibilities for how to model this. This is a model comparison question - you should implement each possibility you think is reasonable as DCMs and use Bayesian Model Selection to compare the evidence. For instance: - If you have bidirectional connections between the driving region, region A, and the secondary region B, then the difference you observe between conditions in region A could be due to the backward connections from region B. So have the union of both conditions drive region A, then have the backward connection modulated by condition B. - You could model the conditions as separate driving inputs entering region A with no modulatory effects. Etc Regarding the interpretation of modulatory inputs. Yes, your interpretation largely makes sense but can be simplified – modulatory inputs are increases or decreases in coupling between regions in the context of an experimental condition. The driving inputs are the same, except they innovate regions directly rather than the connections between regions, and in doing so they stimulate the activity (dynamics) in the system. The model does not speak to the mechanisms of how these effects enter the system. Best, Peter From: Rita Elena Loiotile [mailto:[log in to unmask]] Sent: 11 June 2015 00:44 To: Zeidman, Peter Subject: Re: [SPM] DCM questions Hi Peter, Thank you for the response. I'm sorry, I think I was a bit confusing in my initial email. (Also my conjunction statement was incorrect. Sorry!) My distinction between stimulus and component was meant to illustrate that I was setting up my GLM according to what-- I think-- is appropriate for DCM. Therefore, in the attention to motion example, if we think about the 2 stimuli photic and motion, the GLM is not constructed by entering each stimulus as a separate regressor, but rather by entering the conditions photic (photic stimulus U motion stimulus) and motion (motion stimulus). What this means is that a t-test for the main effect of the motion component in the GLM for DCM ends up being roughly equivalent to a t-test of the contrast "motion stimulus - photic stimulus" in a non-DCM GLM. That's correct, no? So then my question is: If, for example, V5 were active for both the photic and motion contrasts, and I used photic as a driving input to V1, would I have to separately modulate both the photic and motion connectivity to V5? And, if so, could I do a t-test or something on the modulatory connectivity parameters to compare which condition-- photic or motion-- produces greater modulation of the afferent V5 connection? I hope that's a bit clearer. For the second question: I think I understand your response. I guess a more pointed way to ask my original question would be: what is the difference in interpretation between a model that has a motion driving input to V5 and another model that has motion modulating an afferent connection to V5? I'm not sure what "experimental and from outside the brain" really means because ultimately driving inputs and modulatory connections are both driven by an exogenous condition and get into the brain somehow. My guess is that the first model (with the driving input) signifies that there is modulation of an afferent connection to V5, but the afferent connection comes from an area of the brain that has not been included in the model space. Is this correct? Thanks again, Rita On Wed, Jun 10, 2015 at 10:45 AM, Zeidman, Peter <[log in to unmask]<mailto:[log in to unmask]>> wrote: Hi Rita, I’m having a little difficulty following, perhaps you could clarify how you have set this up. First, when you talk about 2 components, do you mean 2 sets of voxels identified by 2 contrasts? Assuming that’s right... If I understand correctly, you are surprised that you get activations for both contrasts. A decision on how to set up the DCM will rest on understanding what these contrasts show. Your first contrast involves a union operator – did you implement this with a conjunction test? Which type? Regarding your second question. In DCM, you define the boundaries of the system, through the regions you select. Dynamics in the system are caused by the driving inputs, which will generally be experimental and from outside the brain. The bilinear term (the modulatory inputs) relate experimental inputs to changes in coupling strength, which is analogous to the probability of transmission changing with the experimental condition or context. If, however, you want to model the change in coupling as depending on the activity in a third region, rather than just an experimental context, you can represent this with an additional non-linear term. See the original non-linear DCM paper (Stefan et al. 2008, NeuroImage) for further discussion on the distinction between bilinear and nonlinear terms. Best, Peter From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]<mailto:[log in to unmask]>] On Behalf Of Rita Elena Loiotile Sent: 09 June 2015 23:27 To: [log in to unmask]<mailto:[log in to unmask]> Subject: [SPM] DCM questions Hi, I am interested in doing a DCM analysis and have 2 conceptual questions I'm hoping someone can help me with. Sorry that this is so long! 1) I have 2 stimuli: Stim1 and Stim2. I will refer to the common component between Stim1 and Stim2 as Component1 (i.e. Stim1 U (Stim2 - Stim1)) and the component unique to Stim2 (i.e. Stim2 - Stim1) as Component2. Component2 significantly activates region A. At first, I thought this would be a straightforward example in which I put a driving input of Component1 in some regions that are active under Component1 and then test modulatory connectivity to region A for Component2. This is analogous to assuming a photic driving input in V1 and testing modulation of afferent connections to V5 for motion stimuli. The part that is not analogous is that, while photic input doesn't significantly activate V5 on its own, I do see a weak but significant activation of Component1 on region A. Therefore, should I set up my model to have a driving input of Component1 on Component1-activated regions and then modulatory connectivity to region A for both Component1 and Component2? My thinking is that if I only specify modulatory connectivity for Component2, it will incorrectly subsume part of the Component1 modulatory connectivity. If so, can you please suggest what statistical tests I may be able to perform to show that Component2 modulatory connectivity is significantly greater than Component1 modulatory connectivity? (A paired t-test?) 2) Several DCM papers analogize driving and modulatory DCM inputs to driving and modulatory neuronal afferents. According to Sherman & Guillery, "The driver can be identified as the transmitter of receptive field properties; the modulator can be identified as altering the probability of certain aspects of that transmission." Maybe I am misunderstanding this, but it seems like, then, that the proper analogs to driver and modulator neurons are, respectively, bilinear and nonlinear connections. But, if that's the case (which it may not be if I've just totally misunderstood this), then what exactly would driving inputs represent? To me, it seems like driving inputs represent input from areas that have not been included in the modeled system. For example, driving input to V1 occurs in the attention to motion example because V1's actual inputs have been left out of the model (if they were included, it would probably have been evidenced as a modulation of connectivity between V1 and its inputs). Okay. That's it for now. Thanks again, Rita