Effects of ketamine on rat social behavior as analyzed by DeepLabCut and SimBA deep learning algorithms

Abstract

Traditional methods of rat social behavior assessment are extremely time-consuming and susceptible to the subjective biases. In contrast, novel digital techniques allow for rapid and objective measurements. This study sought to assess the feasibility of implementing a digital workflow to compare the effects of (R,S)-ketamine and a veterinary ketamine preparation Vetoquinol (both at 20 mg/kg) on the social behaviors of rat pairs. Historical and novel videos were used to train the DeepLabCut neural network. The numerical data generated by DeepLabCut from 14 video samples, representing various body parts in time and space were subjected to the Simple Behavioral Analysis (SimBA) toolkit, to build classifiers for 12 distinct social and non-social behaviors. To validate the workflow, previously annotated by the trained observer historical videos were analyzed with SimBA classifiers, and regression analysis of the total time of social interactions yielded R ² = 0.75, slope 1.04; p < 0.001 (N = 101). Remarkable similarities between human and computer annotations allowed for using the digital workflow to analyze 24 novel videos of rats treated with vehicle and ketamine preparations. Digital workflow revealed similarities in the reduction of social behavior by both compounds, and no substantial differences between them. However, the digital workflow also demonstrated ketamine-induced increases in self-grooming, increased transitions from social contacts to self-grooming, and no effects on adjacent lying time. This study confirms and extends the utility of deep learning in analyzing rat social behavior and highlights its efficiency and objectivity. It provides a faster and objective alternative to human workflow.

Keywords: DeepLabCut; EQIPD quality system; NMDA receptor antagonist; SimBA; ethology.

FIGURE 1

Presented are the dates, sources and numbers of videos used for DeepLabCut network training (upper), for DeepLabCut pose-estimation’s validation and SimBA’s classifiers’ building (middle), as well as for the statistical analyses of the historic videos (bottom left) and for the novel videos’ analysis (bottom right). Blue color indicates Vetoquinol ketamine data inclusion while the orange color shows videos of rats treated with (R, S)-ketamine. The pink color shows videos of rats not treated with ketamine. MH and DP are the initials of the highly trained observers.

FIGURE 2

Regression analyses of 101 unpublished historical videos with rat social behavior analyzed in our laboratory over the years 2017–2018 by the experienced researcher (MH) and analyzed by a digital workflow (DeepLabCut and SimBA toolkits). The axes represent time of social behavior in seconds. The animals were treated with vehicle, Vetoquinol ketamine and other compounds. All specific behavioral categories (A–G) as well as their sum (H), yielded significant R ² values (see Results). While fighting’s (E) and grooming’s (F) regression coefficients were significant (p < 0.001), it was apparent that the human has classified fighting-like behavior less frequently than SimBA, and grooming-like behavior more frequently than SimBA. Bearing similarities between these two behaviors (mostly conspecific’s fur chewing), likely difficult to distinguish for the human and SimBA, we decided to combine them into a fighting + grooming category (G), which yielded R ² = 0.67, slope of 0.83 and p < 0.001.

FIGURE 3

Statistical comparisons of historical videos with rat social behavior analyzed over the years 2017–2018 by MH (left panels) and by the digital workflow (DeepLabCut and SimBA toolkits; right panels). Nineteen videos of vehicle- and 19 videos of Vetoquinol ketamine-treated rats were included and their analyses were compared side-by-side. Panels (A,E) show the sum of time in seconds of social behaviors, typically presented in literature. Analyses of specific behavioral categories (B,F) demonstrate remarkable similarities between human and digital workflow’s as the post hoc analyses revealed decreases (p < 0.05–0.001) in all 7 behaviors analyzed in both datasets, except mounting + crawling, for which SimBA found no difference between Vetoquinol ketamine and vehicle groups. Treatment with ketamine appeared to reduce the percentage of fighting and following behaviors [(C,D) and (G,H), respectively] as assessed by the human and digital workflows. Additional analyses with digital workflow only revealed a decrease of movement in rats treated with Vetoquinol ketamine (I), being observed for the whole time of observation (J), insignificant increase of adjacent lying (K), significant increase of immobility (L) and self-grooming time (N) but no effect on rearing time (M). Panel (O) shows increased coefficients of self-grooming following social contacts derived from FSTTC analysis. The images on the bottom left show 12 representative frames of behavioral categories analyzed by SimBA toolkit (an example of immobility being marked as Drug Side Effect). Symbols: Veh: vehicle, Ket: Vetoquinol ketamine. Asterisks and numerical values above bars indicate differences (P’s) toward vehicle conditions. Data distributed normally are shown as mean ± SEM; data lacking normal distribution are shown as the violin plots.

FIGURE 4

Effects of vehicle, Vetoquinol ketamine and pure (R,S)-ketamine (both at 20 mg/kg, 30 min before the test) on rat social behavior as analyzed by the digital workflow (DeepLabCut and SimBA toolkits). (A) shows the sum of time in seconds of social behaviors, typically presented in literature. Analyses of specific categories (B) demonstrate decreases of anogenital sniffing, body sniffing and following behaviors due to the treatment with both compounds, no effect on mounting + crawling and suppression of fighting + grooming induced by (R,S)-ketamine only. Treatment with ketamine preparations appeared to reduce the percentage of the following behavior due to Vetoquinol ketamine (D) and of fighting + grooming due to (R,S)-ketamine treatment (E) as compared to Vehicle (C). Digital workflow revealed also a decrease of movement in rats treated with both ketamine preparations (F), evident for the whole time of observation (G), insignificant increase of adjacent lying (H), and significant increase of self-grooming due to Vetoquinol ketamine (K). There was no effect of either compound on the rearing time (J), while the effect on immobility could not be calculated as only one control rat displayed this behavior (I). Panel (L) shows that both ketamine preparations increased coefficients of self-grooming following social contacts derived from FSTTC analysis. (M–O) show the most prominent (FSTTC cut-off > 0.5) transitions of all behavioral categories and suggest that the most complex was behavior of vehicle treated rats, followed by behavior of (R,S)-ketamine treated animals, while Vetoquinol ketamine treated rats exhibited the least complex behavioral transitions. (P) shows all 7 social contact behaviors (anogenital sniffing, crawling, fighting, grooming, mounting, nosing, and sniffing) leading to self-grooming with FSTTC coefficients > 0. In Vetoquinol ketamine animals (blue) it was mostly fighting, mounting, nosing and crawling, while in (R,S)-ketamine animals (orange) it was fighting and crawling that led to self-grooming. Symbols: Veh: vehicle, Ket: Vetoquinol ketamine, (R,S)-KET: pure (R,S)-ketamine. Asterisks and numerical values above bars indicate differences (P’s) toward vehicle conditions. Data distributed normally are shown as mean ± SEM; data lacking normal distribution are shown as the violin plots.