Optimization of the Ultrasonic Vocalization Classifier to Analyze Mouse Vocalizations
Authors:Tracey Chan, Natalia Malkova
Mentor:Paul Patterson, Anne P. and Benjamin F. Biaggini Professor of Biological Sciences, California Institute of Technology
Mice communicate using ultrasonic vocalizations (USVs), which are sounds with high frequencies that are inaudible to humans. Mouse pups produce USVs during isolation from the mother and littermates, while adults emit USVs primarily when interacting with other adults. An approach to analyzing mouse USV structure is the categorization of USVs into ten distinct syllable types according to their duration and changes in frequency. In collaboration with Guangying Wu and his group, now at George Washington University, we are developing a Matlab-based program, USV Classifier, to efficiently classify syllables and to analyze their sequences in USV sample files. This software will allow for greater understanding of mouse communication, especially syllable preferences and syllable combinations that are associated with various social settings and age groups. To validate the accuracy of USV Classifier, I compare “manual” classifications, which are labeled by researchers, to “automatic” classifications, which are generated by USV Classifier, for five three-minute-long pup USV samples. Software performance in syllable classification and sequence analysis is being improved by modifying source code and adjusting user-set parameters. I have changed certain syllable-detecting code functions to reduce overall average classification error to 18% and missed detection to 3%. I also have added software functions to group closely-emitted syllables together and to output average syllable frequency change and power. The optimal software will next be applied to study USVs in a mouse model for an autism risk factor, maternal immune activation (MIA). The offspring of immune-activated dams produce fewer and different USV syllables than offspring of dams injected with saline, both as pups and adults. The reduction in verbal communication is reminiscent of deficits characteristic of human autistic patients. Syllable use by MIA offspring will be further investigated using an improved version of USV Classifier.