Machine learning tracks fish schooling evolution across thousands of species

Study quantifies schooling gains and losses

Scientists investigated the evolution of collective movement across thousands of fish species using advanced computational methods. The research indicates that independent lineages gained the ability to school approximately 60 times over the last 350 million years. Conversely, these groups lost the capacity to move together as a unit roughly 320 times during their evolutionary history. Andrew Hein served as senior author while Jay Love acted as first author for this large scale project.

Text and image data drive analysis

The team utilized a text transformer method to encode written descriptions of species into numerical data for interpretation. They also analyzed citizen science images from iNaturalist to assess fish direction and alignment in three dimensions. This approach allowed the researchers to quantify phenotypes for 2,839 species representing nearly 10 percent of all living fish. Previously similar studies included only one or two species totaling a few dozen examples before this work.

Losses of behavior are common

Experts previously believed that losing collective movement behavior was rare but the data proves otherwise. Hundreds of examples exist where a lineage engaging in collective movement produced a solitary daughter species. Genetic mutations lead to changes favoring fitness and survival which allow traits to be inherited by descendants. Once a trait becomes no longer beneficial, descendant species can lose it according to the evolutionary tree.