What is familiar and novel in the Visual behaviour and 2P data set

Hi, I am doing a project through Neuromatch Academy using the Visual behaviour and 2P data set.

We were trying to extract familiar versus novel sessions from the original file, which is described here:visual_behavior_compare_across_trial_types. However, the sessions that are novel and familiar aren’t clear.

In the case shown here the distinction appears to be the ‘ophys_experiment_id’. Should we read session_type (e.g. OPHYS_5_images_A) where the number is the session number and the letter indicates familiarity/novel set? Where A = novel and B = familiar?

However, that doesn’t seem to work for every mouse. So would it be better to rely on the number of ‘prior_exposures_to_image_set’?

Does a prior_exposures_to_image_set == 4 imply the mouse has been exposed to the set 5 times and that is not enough to make the images familiar? I noticed some prior_exposures_to_image_set go up to 49, so I gather this is enough to make the images familiar?

Is it also correct that mice are first exposed to omissions in OPHYS_1_images_B? Is 6 sessions with exposures to omission enough to become familiar with them or is that something we’ll have to find out?


Hi Tristan,

The relationships between the image set, session_type, and prior_exposure_to_image_set can definitely be confusing. Part of this confusion arises because there are multiple dataset variants within the larger Visual Behavior dataset. These are indicated by the project_code. Each mouse belongs to one of these project_codes.

In these different variants, some mice train with image set A, which becomes highly familiar over training (project_code = VisualBehavior or VisualBehaviorMultiscope), while other mice train with image set B, which is then the familiar image set in those mice (project_code = VisualBehaviorTask1B). So the image set letter (A vs B) does not correspond directly to familiarity or novelty, it is the training history of the mouse that counts.

The prior_exposures_to_image_set is the ground truth for how many times (behavior or ophys sessions) a mouse has been exposed to a given image set. prior_exposures_to_image_set = 0 indicates a truly novel exposure to an image set. The same logic applies to prior_exposures_to_omissions. A value of 0 indicates the first session where the mouse experienced omissions.

The session_type does not always align with the expected value of prior_exposures_to_image_set or omissions because a given session_type can be acquired multiple times. This is because some sessions are excluded due to our stringent quality control (QC) criteria. When this occurs, that session_type is retaken. This also means that the sessions that passed QC and were included in the final dataset are not always in the exact order indicated by the session_number (OPHYS_1-6).

I think this schematic might help clarify some of these details:

 Yes, that is an excellent scientific question!! How many exposures with omissions, or a given image set, are needed before they become familiar? And what does it mean to become “familiar”? What kinds of behavioral or neural readouts could you use to evaluate whether something has become “familiar”? These types of questions are great beginnings for some interesting projects and discoveries :slight_smile: