The go and catch trial in the Allen Visual Behavior 2P dataset

Hello, I’m trying to use the 2P dataset for my own project, but I have some questions about the experiment design.

As shown in the picture, the “go trial” means the images changed, but I don’t fully understand the “catch trail”. If it means the images have no change, the first several trials in the “go trial” should also means the “catch”; and the end of the “catch trial” is also a “go” because it has a image change.
So I want to know how the define the “go” and “catch” trial. Are they exist in the mice’s behavior experiment or just in the data analysis? Are they distinguish because of the number of non-change images?

I am bringing this up because I found the number of “go trial” and “catch trial” does not meet my expectations when I looked at the performance data. I used to think that every image’s onset means a trial, changed image corresponding to a “go” trial and non-changed image corresponding to a “catch” trial. If it’s true, the “catch” number should much larger than the “go” tiral. So I guess you have already defined the “go” trial and “catch” trial in other ways, and I wonder the specific definition.


Hi Theresa,

Thanks for the question. Here is some additional information about the trial structure in the task:

Each ‘trial’ consists of multiple stimulus presentations, starting with one stimulus being repeatedly presented (the ‘pre-change’ image). At the start of the trial, a stimulus is chosen as the ‘change’ (or ‘sham change’) image, and a ‘change time’ is drawn from a geometric distribution between 2-8 seconds after the trial starts.

On go-trials, the stimulus identity is changed at the selected change time, and we record the animal’s response in a 750ms window after the change to determine whether the trial was a hit or miss, and a reward is delivered if it was a hit. On catch-trials, the stimulus identity is not changed, but we record the animal’s response in a comparable 750ms window after the sham change to evaluate whether the catch trial was a false alarm or a correct rejection. This ensures that the stimulus presentation we use for calculating the false alarm rate on catch trials has the same probability of being a change stimulus as the change stimuli on go trials. This controls for the possibility of animals having learned the underlying temporal distribution. You can see more details of the task design by clicking ‘Download Technical Whitepaper’ at Visual Behavior - 2p -

It’s also worth noting that the number of go and catch trials is not matched, because there are 64 possible pairwise transitions between the 8 images presented in the session, and only 1/8 of those are the same image repeated (aka catch-trials) That explains the larger number of go-trials vs. catch-trials in the example you shared.

It’s also possible to use alternative definitions for catch trials. Our definition (using only non-change stimuli that are perfectly temporally matched to the change stimuli) is the most conservative thing to do. Alternatively, you could treat every non-change stimulus as a catch trial, or apply some additional rule of your own.