Head-twitch response

Detection of head-twitch responses in mice was performed as previously reported (de la Fuente Revenga et al. 2020) with additional computational (Halberstadt 2020) and visual inspection. Neodymium magnets (N50, 3 mm diameter × 1 mm height, 50 mg) were glued with cyanoacrylate to the top surface of aluminum ear tags for rodents (Las Pias Ear Tag, Stoelting Co.) with the magnetic south of the magnet in contact with the tag. Mice were ear-tagged bilaterally through the pinna antihelix under ketamine/xylazine anesthesia (120 mg/kg and 12 mg/kg, respectively) and allowed to recover before conducting behavioral testing. After treatment with DOI (0.5 mg/kg) or vehicle, ear-tagged mice were placed inside a plastic container surrounded by a coil (~500 turns 30 AWG enameled wire) of which output was amplified (Pyle PP444 phono amplifier) and recorded at 1000 Hz using a NI USB-6001 (National Instruments) data acquisition system. Testing lasted 30 min. Data acquisition was performed as previously described (de la Fuente Revenga et al. 2019). Data were processed using a previously described signal analysis protocol (de la Fuente Revenga et al. 2020). To refine head-twitch detection, the signal was also processed using a deep learning-based protocol based on scalograms (Halberstadt 2020). Mismatches between both detection methods were inspected visually without clues relative to the timestamp of the event or treatment group. After manual classification of dubious events, the final values of head-twitch counts were corrected with a custom script. The ∣V∣ threshold for the initial filter using the findpeaks function in MATLAB (Mathworks, R2019a) was halved to a more permissive (0.01 ∣V∣) threshold than previously reported (de la Fuente Revenga et al. 2020). The convolutional neural network (CNN) employed was trained using head-twitch (~1200) and non-head-twitch (~600) events at a 1:1 training and test ratio from our previous studies (de la Fuente Revenga et al. 2020) and unpublished data from experiments conducted on magnet ear-tagged mice. The combined false-positive and false-negative rate of the classifier was less than 3% in the test set.