Profiling a Caenorhabditis elegans behavioral parametric dataset with a supervised K-means clustering algorithm identifies genetic networks regulating locomotion

Abstract

Defining genetic networks underlying animal behavior in a high throughput manner is an important but challenging task that has not yet been achieved for any organism. Using Caenorhabditis elegans, we collected quantitative parametric data related to various aspects of locomotion from wild type and 31 mutant worm strains with single mutations in genes functioning in sensory reception, neurotransmission, G-protein signaling, neuromuscular control or other facets of motor regulation. We applied unsupervised and constrained K-means clustering algorithms to the data and found that the genes that clustered together due to the behavioral similarity of their mutants encoded proteins in the same signaling networks. This approach provides a framework to identify genes and genetic networks underlying worm neuromotor function in a high-throughput manner. A publicly accessible database harboring the visual and quantitative behavioral data collected in this study adds valuable information to the rapidly growing C. elegans databanks that can be employed in a similar context.

Figure 1

Schematic representation of the AQUABN system. Left panel: hardware and software of the AQUABN system. Right panel: workflow chart of the AQUABN software.

Figure 2

Clustering of genes according to the behavioral signature of their genetic mutants. The selected genes segregated into six clusters according to the behavioral signature of their mutants. Seed worm strains are indicated in bold font. While eighteen variants unambiguously fell into a single cluster without supervision, seven variants could fit into two clusters. In these cases, the two involved clusters are connected with lines and the variants are labeled with asterisks. The area of each section and the spatial position of each circular diagram are only for presenting purpose without biological meaning. egl-30 (lf) and egl-30 (gf) are a loss-of-function (md186) or a gain-of-function (js126) allele of egl-30, respectively. We also used two loss-of-function mutations of unc-29 (x29 and e193).

Figure 3

Self-aggregation of worm strains according to the behavioral signature of their genetic mutants using an unsupervised K-means clustering algorithm. The selected genes self-aggregated into five clusters and two isolated genes according to the behavioral signature of their mutants. The area of each section and the spatial position of each circular diagram are only for presenting purposes without biological meaning.

Figure 4

Clustering of genes according to the behavioral signature of their genetic mutants with ACKMCA. The selected genes segregated into six clusters with ACKMCA according to the behavioral signature of their mutants. Seed worm strains are indicated in bold font. The area of each section and the spatial position of each circular diagram are only for presenting purposes without biological meaning.

Figure 5

Effect of redundant/correlated data on supervised clustering genes according to the behavioral signature of their genetic mutants. The selected genes segregated into six clusters with ACKMCA according to the behavioral signature of their mutants. Seed worm strains are indicated in bold font. The area of each section and the spatial position of each circular diagram are only for presenting purposes without biological meaning.

Figure 6

Effect of data size on supervised clustering genes according to the behavioral signature of their genetic mutants. The selected genes segregated into five clusters with ACKMCA according to the behavioral signature of their mutants. Seed worm strains are indicated in bold font. The area of each section and the spatial position of each circular diagram are only for presenting purposes without biological meaning.