Repository logo
 

Nonlinear dynamics and machine learning classification of plant pigment patterns

Abstract

Plants exhibit a variety of vibrant colors that are both beautiful and functional. They owe their reds, purples, and blues to a class of pigments called anthocyanins. Many plants possess spatial variation in their anthocyanin concentration and color, which manifest as diverse patterns on their leaves and flowers. Flower patterns can influence interactions with pollinators, who may have innate preferences for certain patterns and can learn to distinguish between them. Recent work has identified the genes and proteins involved in activation and inhibition of anthocyanin synthesis in some species of Mimulus and showed that their dynamics can be described with a two-component diffusion model. In this thesis, we combine numerical simulations of this model with machine-learning algorithms to classify patterns based on a parameter value that influences the pattern spot size and density. A key challenge is to successfully classify using 2-dimensional spot data, which would permit the classification of real petal data from photos. Our approach makes use of the Voronoi mountain function to construct a 3-dimensional surface from the 2-dimensional data. Classification is very successful with simulated data, and it produces plausible results for real Mimulus petals.

Description

Rights Access

Subject

Citation

Associated Publications