Suitability of GLMsingle for passive viewing design with non-repeated stimuli and short catch trials

[gaozhy]

Hello,

I am considering using GLMsingle for an fMRI dataset and would appreciate some guidance on whether my experimental design aligns with the method's assumptions.

In our experiment, participants passively viewed visual stimuli from two categories of interest. Each stimulus was shown only once (i.e., no repeats), and participants were not required to respond. To assess engagement, we included 6 catch trials per run—brief text prompts instructing participants to press a button (catch trial is not of our interest). These catch trials were response-dependent and typically lasted less than 1 second, whereas the main visual stimuli were presented for 4 seconds, and with jittered interval around 2s.

My understanding is that a key advantage of GLMsingle is its ability to improve the reliability of beta estimates by using repeated trials to derive a data-driven noise model. Given the lack of repeats for my main conditions, I am unsure how to best proceed.

Could the repeated catch trials serve as a "repeated condition" to help GLMsingle estimate a general noise pool? I am concerned that using these trials to inform the denoising of my main visual conditions could be problematic, given how different they are in content, duration, and task demands. Is this a valid concern, or can the algorithm effectively leverage any repeated events.

Thanks very much!

[kendrickkay]

The issue you raise is a fair/sensible one. One idea is that you could use GLMsingle just for the HRF tailoring features and ignore the GLMdenoise and ridge regression features (which do require repeats). Another idea is that as you suggest, I think the algorithm could very well leverage your repeated catch trials. Based on my estimate and without getting into the nitty gritty, I do think that although there are some content/duration/task demands differences, there is a reasonably high chance things will work out well. The duration difference is a little tricky. It may be a reasonable approximation to specify a single duration (like 4 or 2 or 1) that may apply well to all the trial types. (We don't really know what the underlying neural activity duration is, ultimately.) You could inspect the diagonistic figures to try to get a sense of whether things are working well, or you could do your own tests. That's about all I can say without knowing a lot more about the details of the experiment.

…

On Jun 26, 2025, at 1:11 AM, gaozhy ***@***.***> wrote: Hello, I am considering using GLMsingle for an fMRI dataset and would appreciate some guidance on whether my experimental design aligns with the method's assumptions. In our experiment, participants passively viewed visual stimuli from two categories of interest. Each stimulus was shown only once (i.e., no repeats), and participants were not required to respond. To assess engagement, we included 6 catch trials per run—brief text prompts instructing participants to press a button (catch trial is not of our interest). These catch trials were response-dependent and typically lasted less than 1 second, whereas the main visual stimuli were presented for 4 seconds, and with jittered interval around 2s. My understanding is that a key advantage of GLMsingle is its ability to improve the reliability of beta estimates by using repeated trials to derive a data-driven noise model. Given the lack of repeats for my main conditions, I am unsure how to best proceed. Could the repeated catch trials serve as a "repeated condition" to help GLMsingle estimate a general noise pool? I am concerned that using these trials to inform the denoising of my main visual conditions could be problematic, given how different they are in content, duration, and task demands. Is this a valid concern, or can the algorithm effectively leverage any repeated events. Thanks very much! — Reply to this email directly, view it on GitHub <#182>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAU2DEU7FM2BLSVCIOKD6NL3FOFK7AVCNFSM6AAAAACAFGK72OVHI2DSMVQWIX3LMV43ERDJONRXK43TNFXW4OZYGQ4TMNBYG4>. You are receiving this because you are subscribed to this thread.

[gaozhy]

Thanks so much for the quick reply! That makes perfect sense and is incredibly helpful.