Rage against the machine: advancing aggression ethology through machine learning

Abstract

Aggression is a highly conserved behavior and exists along a spectrum from adaptive to maladaptive. Adaptive aggression can serve to protect mates, territory, and resources. Maladaptive aggression, however, can present as escalated and uncontrolled, and can occur comorbid with neuropsychiatric disorders including autism spectrum disorders, post-traumatic stress disorder, and intermittent explosive disorder. Inappropriate aggression seeking is detrimental to both individuals and society, and current treatment options are largely ineffective, or associated with significant side effects (Coccaro et al. 2009; Carlson et al. 2010; Frogley et al. 2012; Khushu and Powney 2016). In the clinical literature, aggression is typically delineated into instrumental, reactive (fight or flight), and appetitive (rewarding) phenotypes. Preclinically, there is a long history of research involving reactive aggression, but a much smaller body of work only in males examining the neurobiology of appetitive aggression. The goal of this dissertation was to further develop preclinical models of appetitive aggression in mice by understanding the different behavioral and whole-brain activation patterns between the sexes, and by directly comparing appetitive and reactive aggression phenotypes. A significant portion of my work in this arena has involved developing a machine learning based platform for high throughput and consistent scoring of aggression behaviors - Simple Behavioral Analysis (SimBA). Importantly, I posit that machine learning based behavioral detection paired with artificial intelligence explainability techniques allows users to objectively quantify and share behavioral classifiers in an RRID-like fashion. Using this platform, I have discovered that while both males and females exhibit reactive aggression, males but not females show appetitive aggression. I examined the neural correlates of this behavioral sex difference using whole-brain c-fos activity mapping, identifying a potential network inhibiting appetitive aggression in females. In males, I further identified the lateral septum as a potential locus of differential control of reactive and appetitive aggression. Ultimately, this dissertation indicates that reactive and appetitive aggression are neurally dissociable processes, with an inhibitory network in females gating appetitive aggression.