Image of the Day: ButterflyNet
Image of the Day: ButterflyNet

Image of the Day: ButterflyNet

Scientists used machine learning to analyze the coevolution of physical traits in butterflies.

Aug 16, 2019
Nicoletta Lanese

ABOVE: Co-mimic pairs from the species Heliconius erato (odd columns) and Heliconius melpomene (even column) sorted by greatest similarity from top left to bottom right.
J HOYAL CUTHILL

A machine learning algorithm revealed that coevolving butterflies “borrow” physical features from each other, such as wing shape and pattern, and use them to generate novel features over time, researchers reported August 14 in Science Advances. The scientists set out to test a model known as Müllerian mimicry, which proposes that species sometimes mimic each other to glean mutual benefits; for instance, a predator might mistake an edible butterfly for a toxic one if the insects look similar, which softens the effect of predation on both species. In the past, it was difficult to quantify phenotypic similarity and thereby test the theory.

“Machine learning is allowing us to enter a new phenomic age, in which we are able to analyse biological phenotypes?—what species actually look like?—at a scale comparable to genomic data,” says coauthor Jennifer Hoyal Cuthill, an evolutionary biologist with joint positions at the University of Cambridge, University of Essex, and Tokyo Institute of Technology, in an announcement. After training their algorithm ButterflyNet on 2,400 photographs of Heliconius butterflies, the researchers found that the insects both closely copy each other and generate brand-new traits from borrowed ones, thus creating biological diversity.

J.F.H. Cuthill et al., “Deep learning on butterfly phenotypes tests evolution’s oldest mathematical model,” doi:10.1126/sciadv.aaw4967, Sci Adv, 2019.

Nicoletta Lanese is an intern at The Scientist. Email her at nlanese@the-scientist.com.