Quantitative Classification and Natural Clustering of Caenorhabditis elegans Behavioral Phenotypes Using Machine Vision
Prof. Joong-Hwan Baek
Abstract
Among the organisms most amenable to the genetic analysis of behavior is the nematode Caenorhabditis elegans (a microscopic worm) since they have simple
nervous system and short generation time. Genetic analysis of nervous system function relies on the rigorous description of animal behaviors. However,
standard methods for classifying the behavioral patterns of mutant C. elegans rely on human observation and are therefore subjective and imprecise. Here we
describe the application of machine learning and image feature extraction techniques to quantitatively define and classify the behavioral patterns of C. elegans
nervous system mutants. We have used an automated tracking and image processing system to obtain measurements of a wide range of morphological and behavioral
features from videos of representative mutant types. By performing principal component analysis using a selected subset of features, we represented the behavioral
patterns of eight mutant types as data clouds distributed in multidimensional feature space. Cluster analysis using the k-means algorithm made it possible to
quantitatively assess the relative similarities between worm types and to identify natural clusters among the data. The patterns of similarity identified in this
study closely paralleled the functional similarities of the mutant gene products, suggesting that the quantitative image features are an effective diagnostic of the mutants’ underlying molecular defects.