Machine learning helps explain evolution of fish schools

Scientists have used new machine learning techniques to investigate a topic no one has ever found a way to study: The evolution of fish moving together in schools across thousands of species.

The new paper, published Nov. 6 in the Proceedings of the National Academy of Sciences, is among the largest studies of the evolution of any behavior ever conducted in any group of species, according to the authors. The research shows that behaviors, which have previously been difficult to quantify, can be studied at a large evolutionary scale.

The study reports that fish species have independently gained the capacity to move together as a unitsome 60 times – and lost it some 320 times – across the evolutionary tree of life for fish over the last 350 million years. Species can gain a trait when genetic mutations lead to changes that favor fitness and survival. Once that trait exists, it can be inherited by many descendant species, which can later lose it should it become no longer beneficial.

“Independently, many lineages of fish went from not engaging in collective movement behavior to schooling,” said Andrew Hein, associate professor of computational biology in the College of Agriculture and Life Sciences (CALS). Hein is senior author of the study; Jay Love, a postdoctoral researcher in Hein’s lab, is first author.

Perhaps what was more surprising is that the behavior was lost so many times, Hein said.

“There were hundreds of examples across the tree where a lineage that engaged in collective movement left a daughter species that was solitary,” he said. “If you had asked me or other evolution of behavior experts before this study, we’d tell you that losses of collective behavior are probably very rare, but that’s simply not true.” While the evolution of collective behaviors – such as fish schooling, birds flying in a flock, or even people coordinating a wave in a stadium during a sports event – has been researched before, those studies, each including one or two species, add up to a few dozen species.

Now, thanks to advances in machine learning, the researchers were able to classify both written descriptions of species behaviors as well as citizen science images of fish from iNaturalist and other sources to help quantify and analyze schooling behavior from an evolutionary perspective. The team trained a machine learning method known as a text transformer, the same method behind large language models, to analyze studies and descriptions of species and encode them as numbers that the researchers could then use to interpret whether a particular species might engage in collective movement.