Van der Maaten L., Hinton G. (2008). Visualizing data using t-sne. J. Mach. Learn. Res. 9, 2579–2605. [ Google Scholar] Jin T., Duan F., Yang Z., Yin S., Chen X., Liu Y., et al.. (2020). Markerless rat behavior quantification with cascade neural network. Front. Neurorobot. 14, 570313. 10.3389/fnbot.2020.570313 [ PMC free article] [ PubMed] [ CrossRef] [ Google Scholar] Levine S., Pastor P., Krizhevsky A., Ibarz J., Quillen D. (2018). Learning hand-eye coordination for robotic grasping with deep learning and large-scale data collection. Int. J. Rob. Res. 37, 421–436. 10.1177/0278364917710318 [ CrossRef] [ Google Scholar] The function Reports(), which summarizes data from several animals, receives as input the lists with DataFrames and the file names, as well as the body part of interest to extract the metrics and, if necessary, an area to calculate interactions: To represent the pattern of object interaction among animal groups, the Heatmap() function can also be used to plot concatenated data, facilitating visual comparison between days, groups, or trials ( Figure 2C).

Nourizonoz A., Zimmermann R., Ho C. L. A., Pellat S., Ormen Y., Prévost-Solié C., et al.. (2020). Etholoop: automated closed-loop neuroethology in naturalistic environments. Nat. Methods 17, 1052–1059. 10.1038/s41592-020-0961-2 [ PubMed] [ CrossRef] [ Google Scholar] To develop the PyRAT, we used datasets from the Edmond and Lily Safra International Institute of Neuroscience. Adult male Wistar rats ( n = 12) were placed in an open field arena (59x59 cm with 45 cm tall walls) for 20 min per day for 3 consecutive days. Twenty-four hours later, animals were exposed to two identical objects presented in the open field arena for 5 min. We analyzed 48 videos recorded from a top-down view perspective with a Microsoft LifeCam camera at a resolution of 640 x 480 pixels at 30 frames per second (FPS). Alongside these experiments, neural data from the dorsal hippocampus were collected. All procedures were in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals and approved by a local Animal Care and Use Committee. Kobak D., Berens P. (2019). The art of using t-sne for single-cell transcriptomics. Nat. Commun. 10, 1–14. 10.1038/s41467-019-13056-x [ PMC free article] [ PubMed] [ CrossRef] [ Google Scholar] Rossato J. I., Gonzalez M. C., Radiske A., Apolinário G., Conde-Ocazionez S., Bevilaqua L. R., et al.. (2019). Pkmζ inhibition disrupts reconsolidation and erases object recognition memory. J. Neurosci. 39, 1828–1841. 10.1523/JNEUROSCI.2270-18.2018 [ PMC free article] [ PubMed] [ CrossRef] [ Google Scholar]

Conflict of Interest

A common task in animal behavior analysis is the identification of distinct behaviors, such as rearing, grooming, nesting, immobility, and left and right turns. To automatically classify behaviors, we used a combination of two unsupervised approaches on each video frame. We used the hierarchical agglomerative clustering algorithm to label the clusters (Lukasová, 1979) and a non-linear technique for dimensionality reduction called t-distributed stochastic neighbor embedding (t-SNE) to visualize the result (Van der Maaten and Hinton, 2008). The input of both algorithms is the distances between labeled body parts. This approach was chosen because the relative distance between body parts is invariant to the animal position in the pixel space. Combining these techniques, we created a map where the distances between the body parts of each frame are transformed into 2D space using t-SNE and the color of each point is determined by the label from hierarchical agglomerative clustering ( Figure 3A). The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Publisher's Note A) Image showing the trajectory of one rat for 120 s based on the snout coordinates. (B) Image showing rat body orientation during the entire object exploration session. (C) Average heatmap during the entire object exploration session. (D) Top: Object interaction across the entire object exploration session; Bottom left: Bar plot showing interaction time with objects A and A'; Bottom right: Bar plot showing the number of interactions with object A and A'. Data are expressed as mean ± SD. The distance metric passed in this function is Ward's distance and defines the threshold above which the clusters will not be merged.

Gonzalez M. C., Rossato J. I., Radiske A., Reis M. P., Cammarota M. (2019). Recognition memory reconsolidation requires hippocampal zif268. Sci. Rep. 9, 1–11. 10.1038/s41598-019-53005-8 [ PMC free article] [ PubMed] [ CrossRef] [ Google Scholar] Deep learning (DL) and computer vision research fields are improving the performance of image, video and audio data processing ( Krizhevsky et al., 2012). The use of these approaches to estimate human and animal pose is increasing rapidly. This new direction stems from several factors, including improved feature extraction, high scalability to data, availability of low-cost hardware designed for DL, and pre-trained models ready for deployment ( Toshev and Szegedy, 2014; Redmon et al., 2016; Ilg et al., 2017; Levine et al., 2018; Nath et al., 2019). TD, BS, and AR designed, wrote, tested the library, and performed the analysis of the examples. RH and MG evaluated the algorithms. TD documented the library. TD, RH, MG, and AR wrote the manuscript. All authors contributed to the article and approved the submitted version. Funding After 1 hour: What's this? Is there actual taste under the layer of kerosene? We are getting somewhere: 5/10.

Data Availability Statement

Shake all ingredients except Sprite. Strain over fresh ice and finish with sprite Caribbean Sling Cosmo To enhance cluster visualization, we optimize the t-SNE hyperparameters according to the heuristics reported in Kobak and Berens (2019). Their approach is based on three steps, (1) the use of Principal Component Analysis (PCA) in t-SNE initialization to preserve the data structure in lower dimensions; (2) set the learning rate as η = n/12, where n is the number of data points (frames); and (3) set the perplexity hyperparameter, which controls the similarity between points and governs their attraction, as n/100. In addition, we implemented three metrics to quantify the quality of the t-SNE output ( Kobak and Berens, 2019), (1) the KNN ( k-nearest neighbors), which quantifies the preservation of the local structure; (2) the KNC ( k-nearest class), which quantifies the preservation of the mesoscale structure; and (3) the CPD ( Spearman correlation between pairwise distances), which quantifies the preservation of the global structure. many more examples like that, so please remember that PyRAT offers 'only' a collection of singular tools, not