Fruit flies, often caught crawling on a browning banana or overripe zucchini, are insects that are obviously pretty different from people. But on the inside, they actually share 75 percent of the disease-causing genes with humans. For decades, the genome of these tiny beings have been a prime subject for scientists to probe at questions surrounding how certain traits are passed down generations. Flies, however, can be tricky to keep track of because they’re tiny and hard for human scientists to tell apart.
That’s why a team of researchers at Tulane University created software called Machine-learning-based Automatic Fly-behavioral Detection and Annotation, or MAFDA, which was described in an article in Science Advances in late June. Their custom-designed system uses a camera to track multiple fruit flies simultaneously, and can identify when a specific fruit fly is hungry, tired, or even singing a serenade to a potential mate. By tracking the traits of individual flies with varying genetic backgrounds, the AI system can see the similarities and differences between them.
“Flies are such an important model in biology. Many of the fundamental discoveries started with the fruit fly—from the genetic basis of chromosomes to radiation and mutations to innate immunity—and this relates to human health,” says corresponding author Wu-Min Deng, professor of biochemistry and molecular biology at Tulane. “We want to use this system to be able to actually identify and quantify the behavior of fruit flies.”
Deng and his team of researchers not only developed a machine-learning system that decreases human error and improves the efficiency of studying the Drosophila melanogaster, but were able to identify a gene called the fruitless gene, or Fru.
This gene, known to control pheromone production, was discovered to also control how flies smell pheromones and other chemical signals released by surrounding fruit flies engaged in mating. The gene can control the same behavioral circuit (when over- or under expressed) from completely separate organs in the body, Deng says.
“The fruitless gene is a master regulator of the neurobehavior of the courtship of flies,” Deng said.
Because this software lets researchers visualize the behavior of lab animals (including mice and fish) across space and time, Jie Sun, a graduate student at Tulane University School of Medicine and an author on the paper, says that it enables them to characterize the behaviors that are normal, and the behaviors that might be associated with disease conditions. “The MAFDA system also allows us to carefully compare different flies and their behavior and see that in other animals,” says Sun.
Scientists can gain inspiration from computer science and incorporate it into other fields like biology, says Saket Navlakha, a professor of computer science at Cold Spring Harbor Laboratory who was not involved in the study. Much of our creativity can come from weaving different fields and skills together.
From monitoring the fruit flies’ leaps, walking, or wing flaps, the innovative AI system can allow “us to annotate social behaviors and digitize them,” says Wenkan Liu, a graduate student at Tulane University School of Medicine. “If we use the cancer fly, for example, we can try to find what’s different between the cancer flies’ social event, interaction [and] social behaviors to normal social behavior.”
This deep-learning tool is also an example of advancing two separate fields: computer science and biology. When animals, people or the environment are studied, we gain new algorithms, says Navlakha. “We are actually learning new computer science from the biology.”
The system could also be applied to drug screenings, and be used to study evolution or bio-computation in the future.
“It’s a new area for us to study,” says Deng. “We are learning new things every day.”
