Researchers from the University of Cambridge, the University of Essex, the Tokyo Institute of Technology and the Museum of Natural History in London have used the machine learning algorithm to check that butterfly species can develop similar wing patterns for mutual benefit. This phenomenon, known as the Millerian mimicry, is considered the oldest mathematical model of evolutionary biology and was introduced less than two decades after Darwin's theory of evolution by natural selection.
The algorithm was trained to measure variations between different subspecies of Heliconius butterflies, from subtle differences in the size, shape, number, position, and color of wings characteristics to wide differences in the main model groups.
This is the first fully automated, objective method to successfully measure overall visual similarity, which can be used extensively to test how species use wings evolution as a means of protection. The results are reported in the journal Science Advances.
Researchers have found that different species of butterflies act as role models and mimics, "borrowing" features from each other and even creating new models.
"We can now apply AI to new fields to make discoveries that simply weren't possible before," said lead author Dr Ennifer Jennifer Fontil of the Department of Earth Sciences in Cambridge. "We wanted to test Miller's theory in the real world: were these species merging with each other and if so how much? "We failed to test mimicry through this evolutionary system before it was difficult to measure how similar two butterflies are."
The Miller's theory of mimicry is named after the German naturalist Fritz Miller, who first proposed the concept in 1878, less than two decades after Charles Darwin published The Origin of Species in 1859. Miller's theory suggested that species mimic each other for mutual benefit. This is also an important case study of the phenomenon of evolutionary convergence, in which the same characteristics develop over and over again in different species.
For example, Miller's theory predicts that two equally bad tastes or toxic populations of butterflies in the same location will resemble each other because they will both benefit by "sharing" the loss of some individuals to predators who learn how bad they taste. . This provides protection through co-operation and interoperability. It runs counter to the myth of Batesia, which proposes harmless species imitate harmful to protect themselves.
Heliconius butterflies are well-known mimics and are considered a classic example of Mullerian mimicry. They are widespread across tropical and subtropical areas of America. There are more than 30 different types of recognizable patterns within the two types the study focused on, and each type of pattern contains a pair of imitated subspecies.
However, because previous studies of wing models had to be done manually, it was not possible to perform large or in-depth analyzes of how these butterflies mimicked each other.
"Machine learning allows us to enter a new phenomena, in which we can analyze biological phenotypes – what species actually look like – on a scale comparable to genomic data," said Jolin Kutil, who also holds positions at the Tokyo Institute of Technology and University. in Essex.
Researchers used more than 2,400 photographs of Heliconius butterflies from the Museum of Natural History collections, representing 38 subspecies, to train their algorithm, called "ButterflyNet."
ButterflyNet was trained to classify photos, first by subspecies, and then to measure the similarity between different wing patterns and colors. He drew different images in a multidimensional space, with more similar butterflies closer together and fewer similar butterflies further away.
"We have found that these species of butterflies borrow from each other, which confirms Miller's hypothesis of mutual co-evolution," said Jolin Kutil. "In fact, the convergence is so strong that they imitate different species more similar to members of the same species."
Researchers have also found that Müllerian mimicry can generate completely new models by combining features from different vines.
"Intuitively, you expected that there would be less winged models where species mimic each other, but we see exactly the opposite, which was an evolutionary mystery," said Jolin Kutil. "Our analysis has shown that mutual co-evolution can actually increase the diversity of patterns we see, explaining how evolutionary convergence can create combinations of new features and add to biodiversity.
"By exploiting AI, we have discovered a new mechanism by which mimicry can produce evolutionary novelty. In contrast, mimicry itself can generate new patterns by exchanging features between species that mimic each other. Thanks to AI, we are now able to measure the extraordinary diversity of life to make such new scientific discoveries: this can open up new ways of research in the natural world. "
Ennifer F. Folih Kutil et al. 'Deep learning of butterfly phenotypes tests the oldest mathematical model of evolution.' Advances in Science (2019). DOI: 10.1126 / sciadv.aaw4967