Overview

Statement of Problem

For this project, our primary data source is mouse behavioral data from the Tecott Lab at UCSF. [1] The lab has recently developed a method for continuous high-resolution behavioral data collection and analysis, which enables them to observe and study the structure of spontaneous patterns of behavior (“Lifestyles”) in the mouse [Tec03][TN04][GSJ+08][AP14]. They have found that using this method: 1) reveals a set of fundamental principles of behavioral organization that have not been previously reported, 2) permits classification by genotype with unprecedented accuracy, and 3) enables fine dissection of behavioral patterns.

Project Goal

The goal of this project is to explore the effects of genetics on behavior in mice and extrapolate these findings to improve treatment of psychiatric diseases in humans.

Computational Ethology

Ethology is the study of animal behavior, and it generally follows one of two approaches. The first approach was developed by B.F. Skinner and relies heavily on laboratory experiments. Skinner postulated that since behavior is predictable, it should be controllable. His most prominent experiment involved training pigeons through operant conditioning. He found that the pigeons had the ability to learn new behaviors under a reward system. However, the setting of the experiment was incredibly artificial and controlled. In contrast, Konrad Lorenz believed that the only way to truly understand animal behavior is to observe them in their natural context as behavior induced by an artificial environment may not reflect animal behavior in their natural environment. Lorenz’s most prominent experiment discovered that a young goose will instinctively bond with the first moving object that it perceives in order to better recognize its own species. This experiment was conducted in a more natural setting to minimize human manipulation.

This project adopts Lorenz’s approach—observing the whole spectrum of animal behavior in their natural state. Previous approaches relied on human observation to score animal behavior. However, slow data collection, low dimensional data, imprecise and subjective measurements, and human visual and language limitations impeded data collection and analysis. Using modern quantitative tools for measurement, description, and analysis, the field of computational ethology has emerged to solve these issues. Together, modern mathematics, engineering, and computer science have the potential to establish a causal relationship between genetics and behavior.

Why Mice?

Although the human genome was first sequenced over a decade ago, the relationship between genetics and behavior is still not well understood. For ethical reasons, genes cannot be systematically manipulated in humans. Therefore, the familiar scientific testing approach must be edited—enter mice. There are several reasons the mouse has become the mammal of choice in human behavioral studies. Chief among them is the fact that “approximately 99% of mouse genes have human counterparts—conversely, mouse versions (orthologs) can be identified for 99% of human genes.” Additionally, the brain organization and behavioral responses of humans and mice display many similarities. For example, both mammals display complex processes like hunger and fear. This allows scientists to more easily identify and track these behaviors. In addition, the extreme similarities of intra-strain mice, the space efficiency of maintaining their caged environments, and the speed of reproduction make mice a logical alternative to testing on humans.

Methodology

The traditional technology to analyze behavior is time-intensive and labor-intensive. For example, recording data requires researchers to track behavior uninterruptedly. This experiment utilized a method for continuous high-resolution behavioral data collection and analysis. The home cage monitoring (HCM) system utilized is a network of photobeam feeding detection, drinking detection, and activity platform sensors that records mouse active states and inactive states automatically and incessantly over 24-hour periods where 1 hour per day was reserved for HCM maintenance. This facilitates objective, multi-dimensional computer tracking, providing a higher degree of accuracy compared to human observation.

In this project, the HCM system tracked 170 mice, representing 16 strains or approximately 94% of the genome, logging 500,000 behavioral events per mouse per day over 12 days of data collection after 5 days of acclimation. The scope of the study was limited to male mice as females tended to display cyclical changes in behavior. Mice within each strain display strong homogeneity and low variability in genetic composition. This is in part due to high levels of inbreeding which decreases the randomness between individual mouse samples. By studying the behaviors of theses 16 different strains – each strain with their own unique genetic makeup – we can begin to understand the importance of genetic makeup on mice behavior. We can then extrapolate to humans thanks to the remarkable genetic similarity between the two species.

Statement of Statistical Problems

This project utilized sophisticated statistical methods to process data, including machine learning algorithms and statistical inference. The whole project can be divided into 6 sub-projects:

• Behavioral Model
• Exploration and Path Diversity
• Dynamics of AS Patterns
• Ultradian and Circadian Analysis
• Classification and Clustering of Mice
• Power Laws and Universality

Glossary

• Active State (AS): The active state in this model is when the mouse is using energy, such as foraging, patrolling, eating, or drinking. Active states are energetically costly and can be dangerous in a natural environment.
• Computational Ethology: The use of mathematics, engineering, and computer science to overcome the difficulties that come from using humans to score animal behavior.
• Ethology: The study of animal behavior, including the phenomenological, causal, genetic, and evolutionary aspects.
• HCM System: The system used in this experiment to track variables of interest. The HCM System included photobeam sensors at the feeding stations, capacity based sensors at the drinking station, and an activity platform for position detection using an (x,y) system.
• Home Environment: The home environment is the cage of each mouse containing a home base, a food station, and a water station.
• Inactive State (IS): The inactive state in this model is when the mouse is in a state of energy conservation, such as sleeping or resting at the home base.
• Operant Conditioning: Altering of behavior through the use of positive reinforcement which is given to the subject after eliciting a desired response.
• Phenotype: The set of observable characteristics of an individual resulting from the interaction of its genotype with the home environment.
• Strain: A strain here is a genetic variant or sub-type of of the more general mouse population.

Data

The data includes two directories, intervals and txy_coords, and a npy file named all_features_mousedays_11bins. The all_features_mousedays_11bins.npy contains a \(9 \times 1921 \times 11\) matrix, which represents 9 features among 1921 mouse days in 11 2 hour bins for a day, the 9 features are:

• Food (F): records the food consumption (g) for a certain mouse day and a certain time bin.
• Water (W): records the water consumption (g) for a certain mouse day and a certain time bin.
• Distance (D): records the movement distance for a certain mouse day and a certain time bin.
• ASProbability (ASP): records the AS time proportion in the certain time bin.
• ASNumbers (ASN): records the numbers of AS in the certain time bin.
• ASDurations (ASD): records the total duration of AS in a certain bin.
• ASFoodIntensity (ASFI): equals F/ASP.
• ASWaterIntensity (ASWI): equals W/ASP.
• MoveASIntensity (ASMI): equals D/ASP.

The intervals directory has 6 sub-directories, all sub-directories have about 33 files for 3 strains, and for each strain there are 11 days data:

• F: records start and stop time of eating behaviors for a certain strain and a certain day.
• W: records start and stop time of drinking behaviors for a certain strain and a certain day.
• AS: records start and stop time of AS for a certain strain and a certain day.
• M_AS: records start and stop time of movements in AS for a certain strain and a certain day.
• IS: records start and stop time of IS for a certain strain and a certain day.
• M_IS: records start and stop time of movements in IS for a certain strain and a certain day.

The txy_coords directory has 5 sub-directories,all sub-directories have about 33 files for 3 strains, and for each strain there are 11 days data:

• CY,CX,CY: records the position (x,y) in time t for a certain strain and a certain day.
• C_idx_HB: indicates whether the mouse is in HB or not at time t.
• recordingStartTimeEndTime: records the start and stop time of tracking (x,y,t) for a certain strain and a certain day.

Footnotes

[1]	http://www.neuroscience.ucsf.edu/neurograd/faculty/tecott.html