Machine learning in the observation of behavioral phenotypes.

Authors

Gabrielle Cohn

Document Type

Article

Publication Date

Summer 2017

JAX Location

In: Student Reports, Summer 2017, Jackson Laboratory

Abstract

Addiction is a chronic health issue which puts great strain on sufferers and their families. Understanding and identifying the genes associated with addiction will lead to more effective treatments.The Kumar lab studies the genetics underlying behaviors like addiction using mouse genetics and genomic methods. Genetic research methods to study the behavior of an organism require precisely quantifying one or more behavioral traits. Traditionally, this quantification has involved a human observer recording the behavioral characteristics of interest for individual animals. However, th is method lacks scalability for genomics studies and, as behavioral characteristics can be subjective, human observation may lack the precision required for high quality research. To address this problem, I have implemented an unsupervised machine learning algorithm for detection of stereotyped behaviors that is precise and scalable. This approach is unbiased and removes the human component from behavioral studies by using computer vision techniques. It is also a highly scalable pipeline that can analyze thousands of hours of video using cluster based compute such as the JAX.-Cadillac resource. I benchmarked thi method on a collection of videos containing mice with behavioral deficits. My goal was to automatically extract behavioral modalities that differ between control and knockout mice.

Please contact the Joan Staats Library for information regarding this document.

COinS