Action detection using a neural network elucidates the genetics of mouse grooming behavior

Abstract

Automated detection of complex animal behaviors remains a challenging problem in neuroscience, particularly for behaviors that consist of disparate sequential motions. Grooming, a prototypical stereotyped behavior, is often used as an endophenotype in psychiatric genetics. Using mouse grooming behavior as an example, we develop a general purpose neural network architecture capable of dynamic action detection at human observer-level performance and operate across dozens of mouse strains with high visual diversity. We provide insights into the amount of human annotated training data that are needed to achieve such performance. We survey grooming behavior in the open field in 2500 mice across 62 strains, determine its heritable components, conduct GWAS to outline its genetic architecture, and perform PheWAS to link human psychiatric traits through shared underlying genetics. Our general machine learning solution that automatically classifies complex behaviors in large datasets will facilitate systematic studies of mechanisms underlying these behaviors.

Gait-level analysis of mouse open field behavior using deep learning-based pose estimation

“…Top down videos have routinely been used in behavioral neurogenetics and this method could be applied to archival video data. Indeed, we analyzed gait in a strain survey data set that we partially analyzed for tracking and grooming behaviors [42,99]. Because our method does not require expensive specialized equipment, we can also allow the mouse time to acclimate to the open field and collect data over long periods of time, potentially over days with the addition of bedding, food, and water.…”

“…All behavioral protocols have been previously published [42,100]. The strain survey data was published before and reanalyzed for gait behavior [42,99]. We excluded animals that had too few strides which disproportionately affected low activity strains, leading to the final animal numbers indicted in Table 4.…”

“…The R package mousegwas, previously described in [99], was used to compute genome wide associations. Briefly, the classical mouse strains were used (excluding wild-derived strains), ten individuals from each strain and sex combination were randomly selected and GEMMA was used with its LMM method.…”

A machine vision based frailty index for mice

“…These frame-by-frame measurements were used to calculate a variety of pervideo features. These features included traditional open field measures such as anxiety, hyperactivity [23], neural network-based grooming [24], and novel gait measures [25]. All features are defined in Section 3.2 and Supplementary Table S1.…”

“…We along with a number of other groups have applied these new methods to animal behavior analysis. Our group has developed methods for image segmentation and tracking in complex environments [23], action detection [24] and pose-based gait and whole body coordination measurements in the open field [25]. These and other highly sensitive methods have advanced animal behavior extraction [15,26,27].…”