Machine Learning Approaches Reveal Prominent Behavioral Alterations and Cognitive Dysfunction in a Humanized Alzheimer Model

Behavioral manifestations define neurological disorders, but our knowledge of disease-induced behavioral alterations is incomplete and largely limited to standard domain-specific behavioral approaches. New humanized App knock-in (KI) models of Alzheimer’s disease (AD) mimic disease mechanisms better than transgenic overexpression, but fail to display consistent behavioral alterations in standard behavioral tests despite severe AD-related neuropathological changes. To address this technological barrier in AD modeling, we used the machine learning platforms DeepLabCut and VAME to test the hypothesis that deconstructing full sequences of spontaneous mouse behavior into canonical behavioral units (motifs) will better capture AD-induced brain dysfunction in App-KI mice. Indeed, we found that humanized AppNL-G-F/NL-G-F mice have robust impairments in spontaneous behavior as evidenced by prominent alterations in motif use and motif transitions. This phenotype is consistent with cognitive dysfunction, including blunted novelty response, impaired habituation and sensitization, and disorganized behavioral sequences. We conclude that machine learning approaches provide a direct and unbiased measure of AD-related mechanisms of brain dysfunction.