Séminaire de Moritz Lürig

“Computer vision in evolutionary ecology: toward assembling a phenome”

In order to understand biological phenomena related to the phenotype, most ecologists and evolutionary biologists confine their efforts to a small set of observable organismal traits. The century old approach of reducing the complexity of living beings to a tractable number of dimensions has been indispensable in understanding processes such as community assembly, population dynamics and demography, and mechanisms of divergence and adaptation. However, it is becoming increasingly clear that by studying only a small number of traits, often discrete and only limited in space and time, we will not be able to derive causal links between genotypes, environmental factors, and phenotypes. In other words: to construct the genotype-phenotype map of an organism, we need to assemble its phenome. This will allow us to answer the most compelling questions in biological research – for instance, how multivariate selection, phenotypic integration, or pleiotropy affect the strength and direction of evolutionary diversification. Collecting such high-dimensional phenotypic data is typically challenging, time consuming and expensive. Computer vision, the automatic extraction of meaningful information from digital images, is a promising set of methods that may help to alleviate this methodological bottleneck and allow us to collect phenotypic information on a massive scale. The field has blossomed in recent years, and biologists now have a diverse array of open source computational tools at their disposal to assemble phenomic datasets with high throughput and reproducibility. Here I attempt to provide an overview of common automatic and semi-automatic computer vision techniques for the extraction of phenotypic data from digital images.