Automatic quantification of disgust reactions in mice using machine learning

Disgust, a primary negative emotion, plays a vital role in protecting organisms from intoxication and infection. In rodents, this emotion has been quantified by measuring the specific reactions elicited by exposure to unpleasant tastes. These reactions were captured on video and manually analyzed, a process that required considerable time and effort. Here we developed a method to automatically count disgust reactions in mice by using machine learning. The disgust reactions were automatically tracked using DeepLabCut as the coordinates of the nose and both front and rear paws. The automated tracking data were split into test and training data, and the latter were combined with manually labeled data on whether a disgust reaction was present and, if so, which type of disgust reaction was present. Then, a random forest classifier was constructed, and the performance of the classifier was evaluated in the test dataset. The total number of disgust reactions estimated by the classifier highly correlated with those counted manually (Pearson's r = 0.97). The present method will decrease the time and effort required to analyze disgust reactions, thus facilitating the implementation of the taste reactivity test in large-scale screening and long-term experiments that necessitate quantifying a substantial number of disgust reactions.