What can a Worm Learn? Nonlinear Categorization by Caenorhabditis elegans using a Hydrochloric Acid Gradient
Year: 2021
Type: article
Abstract: Spring Valley High SchoolMicroorganisms have been utilized to study the human nervous system for decades.However, the computational capabilities of the neural networks of microorganisms are often not well understood.The goal of this experiment was to determine the level of neural complexity in the microorganism, Caenorhabditis elegans, by examining whether its neural networks are capable of learning categories that are not linearly separable.C. elegans were once thought to possess only a fixed neural network, but then their ability to learn associations and simple linear classifications was discovered.No microorganism is known to have the next level of computational capability, that of performing nonlinear classification.Here, it was hypothesized that C. elegans would be able to learn nonlinearly separable classes.This was tested by setting up three levels of an HCl gradient and associating them with food, in a one-dimensional nonlinear classification problem.Positive and negative associations between HCl levels and the presence/absence of food could only be learned through a nonlinear decision boundary.C. elegans were put into petri dishes with three distinct HCl concentrations of 100μM, 200μM, and 300μM.During training, 100μM and 300μM values were associated with E. coli (positive category), while 200μM was associated with the absence of E. coli (negative category).The learning was then tested in separate dishes by examining the movement of C. elegans towards positively or negatively trained concentrations.In a second experiment, the reverse association (100μM and 300μM negative; 200μM positive) was also established through training and tested.The results showed that C. elegans were able to learn these positive and negative associations, thus acquiring a nonlinear decision boundary (100/300 positive test: x-squared = 24.06,p < 0.001; 100/300 negative test: x-squared = 7.5958, p < 0.023).This demonstrates, for the first time, that the neural networks of C. elegans possess sufficient computational complexity, requiring at least three layers of neurons, to learn nonlinear classification.This advances our understanding of the computational sophistication of the neural networks of even the simplest microorganisms.
Author: Parth Desai
Language: English
Subfield: Cellular and Molecular Neuroscience
Field: Neuroscience
Domain: Life Sciences
Open Access status: green
Funder: University of South Carolina
OpenAlex ID: W3135419571