Emotions are one of the most intriguing products of brain function, yet, we still lack a mechanistic understanding of how emotions arise in the brain [1]. One of the major hindrances to studying the neuronal mechanisms of emotion lies in the difficulty to link observable behavior with internal states and ongoing neural activity [1].

In a recent study, we employed machine-vision and -learning approaches to investigate facial expressions of emotions and their neuronal correlates in mice [2]. Our work built on prior evidence that mice use their orofacial musculature to react to sensory stimuli such as tastants [3] or pain [4]. To assess facial expressions objectively and to detect previously uncharacterized changes, we recorded close-up videos of head-fixed mice and extracted features from video frames using “histograms of oriented gradients” (HOG), a machine-vision technique that represents the statistics of local image features. This allowed us to compare facial expressions of mice reacting to emotion events quantitatively in an unsupervised manner using hierarchical clustering. We next asked whether we could separate facial expressions exhibited upon diverse emotionally salient events into distinct categories by reducing the dimensionality of our data via principal component analysis (PCA) followed by t-distributed stochastic neighbor embedding (t-SNE). This approach separated facial expressions into discrete emotions.

Log in or create a free account to read this content

Gain free access to this article, as well as selected content from this journal and more on nature.com

Access through your institution

or

Sign in or create an account
Continue with Google
Continue with ORCiD