Maladaptive aggression characterizes - or is comorbid with - many neuropsychiatric illnesses, and can have devastating effects on individuals, their caretakers, and healthcare professionals. Human aggression is typically demarcated as exhibiting either reactive (defensive) or appetitive (rewarding) components. Despite a significant clinical awareness of the differences between these aggression presentations, preclinical characterization of their relative circuitry and associated neuronal mechanisms are absent. Using recently established protocols within our lab, we are able to study and compare these aggression phenotypes in outbred male mice in a high throughput manner. Briefly, for appetitive aggression, we train mice to self-administer a novel subordinate intruder over 7 days using a trial design. In the reactive condition, we non-contingently administered intruders with the same frequency distribution as the appetitive mice. In the current experiment, we used CD1xVgat-Cre or CD1xVglut1-Cre mice injected with pGP-AAV-syn-FLEX-jGCaMP7s in the lateral septum (LS) to examine cell-type specific activity via fiber photometry. GABAergic activity in the lateral septum has historically been implicated in the control of reactive aggression, but little is known about the role of excitatory activity in the LS in reactive or appetitive aggression. My roles in this project have included behavioral testing and filming of the mice, as well as scoring these videos for first attacks following intruder presentation. Using these timestamps, I will next analyze the changes in population level dynamics across different time points of aggression motivation, seeking, and consumption using the open source photometry analysis program guPPY. We expect that the photometry results for mice in reactive and appetitive environments will show different patterns of activity, with more glutamatergic activity in the appetitive group, and more GABAergic activity in the reactive groups. I hope to help understand and prevent unnecessary aggression through this research.

Rigorous ethological observation via machine learning techniques, termed computational neuroethology, is a rapidly expanding field. Our lab has created an open-source pipeline for automated behavioral analysis using supervised machine learning called Simple Behavioral Analysis (SimBA), to aid in the high throughput analysis of social behavior. Using pose estimation data of socially interacting animals obtained through open source pipelines such as SLEAP or DeepLabCut, we are able to create large training sets of video frames that are hand scored as positive or negative for a behavior, which we then feed into supervised random forest algorithms. These algorithms then build classifiers which can detect the behaviors in novel videos. My work has focused on building and titrating classifiers for two important social behaviors: face and body sniffing by a dominant mouse toward a subordinate. So far, I have hand-scored a large dataset of social interaction videos to create a sizable training set. I have begun the initial phases of training my classifiers, which involves finding appropriate hyperparameters for the random forest algorithms so that they can differentiate positive and negative behaviors, and refrain from overfitting to our training datasets. Using both machine learning performance metrics as well as hand versus machine comparisons, I am able to understand the generalizability and accuracy of my classifiers. As I continue with this project, I will selectively add more positive and negative examples to correct false positives and boost the confidence of the classifiers through subsequent iterations. This work allows me to gain an understanding of the principles of machine learning techniques, and create classifiers that we openly provide to behavioral neuroscience labs across the world. We expect that the pooling of these classifiers with outside labs will promote a high level of standardization of behavioral definitions in behavioral neuroscience, ultimately increasing reliability and reproducibility.