Kimura, Kotaro et al. (2017) International Worm Meeting "Pan-behavioromics: comprehensive analysis of worms' behavior using machine learning methods and its application to animals in general."

Yamazaki, Shuhei, Takeuchi, Ichiro, Kimura, Kotaro, Maekawa, Takuya

[International Worm Meeting, 2017]

A brain activity can be measured as a large number of neural activities using simultaneous optical monitoring, which is described as a time series vector data of many neural activities. Behavior, the final output of neural activities, however, is still poorly described, often just with one or a few parameters chosen subjectively. This huge asymmetry in data of neural activity and behavior is caused by the difficulty in analysis of behavior: Which aspects of a behavior such as speed, direction, their duration and changing rate, are related to animal's brain activities like sensory perception, memory, and decision-making? Behavioral analyses using machine learning have been reported to classify animal's behavioral states or behavioral patterns (e.g., Brown et al., PNAS 2013), although they have not yet provided clues to understand the causal relationship between stimuli and behavioral response. To explore new methods for a comprehensive analysis of animals' behavioral response to environmental stimuli in collaboration with data science, we applied 3 classic and cutting-edge machine learning methods-decision tree, deep neural network (DNN), and pattern mining-on worms' repulsive odor learning (Kimura et al., J Neurosci 2010). (1) In decision tree analysis, researchers specify a number of features, and the algorithm chooses the ones that effectively classify two groups. We found that naive worms stop straight migration upon sensing a slight increase in the repulsive odor concentration (dC/dt > 0), although learned worms do not respond to it and continue the migration, suggesting that they ignore the "yellow signal". In addition, multiple mutant strains were classified based on the characteristic patterns of component change. (See Yamazaki et al. this meeting.) (2) Using DNN for time series data analysis, we found that, although the speed of naive worms is relatively constant, that of learned worms changed more vigorously in windows of tens of seconds even though the average speeds were the same. The result may suggest that the learned worms avoid the odor using "accelerator and brake" more effectively. (See Maekawa et al. this meeting.) (3) In discriminative pattern mining, we represented odor concentration change and behavioral responses per second as 30 combinations, and found prominently discriminative behavior patterns between wild-type and mutant worms from an extremely large number of possible patterns by using a new sparse estimation technique. In summary, each machine learning method efficiently provided novel and critical knowledge that cannot be obtained from traditional analyses. Such objective comprehensive study of behavioral response, or "behavioromics," allows us to concentrate on key features of behavior. Currently, we intend to apply the above methods for the behavioral analysis of other animals, such as long-travelling sea birds, disease model mice, and humans.

Karla Opperman et al. (2007) International Worm Meeting "A structure-function analysis of the C. elegans L1CAMs."

Lihsia Chen, Xuelin Wang, Karla Opperman, Shan Zhou

[International Worm Meeting, 2007]

Cell adhesion molecules play important roles during development and in maintaining tissue integrity. L1CAMs are a family of immunoglobulin (Ig)-like cell adhesion molecules that are important in vertebrate nervous system development. L1CAMs are conserved in C. elegans, and are encoded by the lad-1/sax-7 and lad-2 genes. SAX-7 and LAD-2 have distinct functions in the nervous system despite overlapping neuronal expression. SAX-7 is required to maintain neuronal positioning while LAD-2 participates in axon pathfinding. We are performing a structure-function analysis to determine how both proteins mediate their functions. Because the cytoplasmic tails between both proteins are distinct, we are focusing our analysis on how the cytoplasmic tail regulates L1CAM functions. Indeed, while the extracellular domains of SAX-7 and LAD-2 are conserved, the LAD-2 cytoplasmic tail is completely divergent and does not contain any of motifs conserved in SAX-7 and vertebrate L1CAMs. These motifs include the ankyrin-binding motif, which links L1CAMs to the spectrin-actin cytoskeleton, the PDZ-binding motif, and conserved tyrosine residues that are phosphorylated. Our preliminary results reveal that both conserved motifs, as well as phosphorylation one of the tyrosine residues, which regulates ankyrin binding, all contribute to the function of SAX-7 in neuronal position maintenance. (1) Chen L, Ong B, Bennett V. J Cell Biol. 2001 Aug 20;154(4):841-55. (2) Wang X, Kweon J, Larson S, Chen L. Dev Biol. 2005 Aug 15;284(2):273-91. (3) Sasakura H, Inada H, Kuhara A, Fusaoka E, Takemoto D, Takeuchi K, Mori I. EMBO J. 2005 Apr 6;24(7):1477-88. Epub 2005 Mar 17.