Quantifying yeast colony morphologies with feature engineering from time-lapse photography

TL;DR

This paper presents a method to quantify yeast colony morphologies over time using texture features from time-lapse images, enabling classification of growth patterns through clustering.

Contribution

It introduces a texture-based feature engineering approach combined with hierarchical clustering to analyze yeast colony development from time-lapse photography.

Findings

Texture scores evolve smoothly during growth

Clustering reveals distinct morphological development trajectories

Hierarchical clustering provides interpretable colony groupings

Abstract

Baker's yeast (Saccharomyces cerevisiae) is a model organism for studying the morphology that emerges at the scale of multi-cell colonies. To look at how morphology develops, we collect a dataset of time-lapse photographs of the growth of different strains of S. cerevisiae. We discuss the general statistical challenges that arise when using time-lapse photographs to extract time-dependent features. In particular, we show how texture-based feature engineering and representative clustering can be successfully applied to categorize the development of yeast colony morphology using our dataset. The local binary pattern (LBP) from image processing is used to score the surface texture of colonies. This texture score develops along a smooth trajectory during growth. The path taken depends on how the morphology emerges. A hierarchical clustering of the colonies is performed according to their texture development trajectories. The clustering method is designed for practical interpretability; it obtains the best representative colony image for any hierarchical sub-cluster.