Social Analysis

[jkbhagatio]

@anayapouget @JaerongA @ttngu207 Some questions to consider for current 24/7 Social experiments. Also see discussion here: SainsburyWellcomeCentre/aeon_experiments#138

In a noisy and changing environment, do mice:

1. Establish a social hierarchy?
   a) Change ranks in the social hierarchy?
   b) Gain resources from foraging consistent with their ranking in a social hierarchy?
2. Exhibit social behaviour?
   a) Interact more or less than expected?
   b) Exert more or less energy / power than expected?
3. Share information that maximises their resources gained from foraging?
   a) Learn to explore until finding the best patch?
4. Compete over resources gained from foraging?
5. Learn to cooperate to maximise resources gained from foraging?

---

1. Tube tests to measure social hierarchy.

• Given this, have an ROI of the corridor, and ask "when animal noses are close together (within some threshold), are they in the corridor?" -> This would count as a tube test event. We should test this with old social data.
• Get all these timestamps, for some window after each of these timestamps, we can e.g. see which animal moves more and which animal's head direction changes (this could imply the animal turned around) - this would be the more submissive animal (and e.g. can rank this by total distance moved within the time window).
• Keep a "Tube Test" table: each row in the table could be a "tube test event", and would specify:
  • which animal were involved
  • metrics of outcome (e.g. as mentioned above: distance moved, head direction change, etc.)
  • Could have some formula that computes "dominance score(?)" given above metrics
    • Make sure dominance score columns are updated for each tube test event
    • Have a rank column also - this could be a tuple (1:N) (this obviously gets trickier with more than 2 animals - transitivity may not hold, but can still use "dominance score" for this)
    • Or could be a rank matrix, where you have the rank for each possible pair
    • Could have a directed graph depiciting current avg of each dominance score between all pairs

2. Look at position data from individual sessions and social sessions.
   • Primary keys for position data table should include animal ID, so that we don't have duplicate timestamp index issues.
     a. Create "virtual" social sessions from individual sessions, compare this to true social sessions.
     b. Look at total mouse movement, energy and power expenditure in social vs. individual sessions.

Have to think more about points 3 and 4 for current 24/7 social experiments. Will worry about point 5 later.

[jkbhagatio]

Some points for 3:

• Measures of learning:

  • Ratio of pellets / time at high patch
  • Stickiness to high patch / patch departure behavior
  • How quickly does one mouse learn as a function of how much time the other is in the high patch?

• Social interaction considerations

  • Social stickiness
  • How to disambiguate visual vs. auditory vs. olfactory cues

• Future considerations:

  • To assess impact of learning from an "expert": Naive mouse with expert vs. Naive mouse alone vs. Two naive mice

[jkbhagatio]

We discussed the following at the DJ Joint Design Meeting on Nov 16:

• Treating blocks as visits
• Full pose model not grrrreat, but probably good enough to get heading / orientation as is
  • Could probably be improved with training on new lower exposure-time videos

Example analyses

1. Social hierarchy

• Example viz / analysis
  • Plot all stats for all tube test events for all subjects
  • Timeseries of dominance ranking per subject
• Example data structures
  • Table of tube test events, where each row is an event, and columns are different features (e.g. duration, winner, score, etc.)

2. Social behavior (including 3. information sharing and 4. competing)

• Example viz / analysis
  • Ethograms of individual and social (centroids less than e.g. centroids < 5 cm apart) behavior
  • Nest time
    • Sleeping time
    • Nest building time
    • Other time
  • Corridor time
  • Arena time
    • Foraging time
      • Patch competing time (probably requires full pose model)
      • Observation time (probably requires full pose model)
    • Other time
  • Example data structures
    • Ethogram table
      • I think ideally want to do something like, per subject, classify behavior discretely within small time windows (e.g. 1 s)

5. Environment stats

• Example viz / analysis
  • Duration, pellets obtained, and wheel distance spun per distribution
  • Duration, pellets obtained, and wheel distance spun on each patch as a function of each distribution
• Example data structures
  • Block tables
  • Patch and Distribution tables that contain one row for each pellet obtained, and columns with properties (e.g. delivery timestamp, wheel threshold, etc.)