Research into animal behavior is currently still dependent on human observations that are not very standardized. But now researchers have developed a computer program that could take over this behavioral observation in the future. It is able to detect animal movements in different situations automatically and in much more detail than a human. As a result, the open source program could in future help standardize behavioral observation and perhaps even uncover completely new behavioral patterns.
Since Jane Goodall researched chimpanzees in the Tanzanian jungle in the 1960s or Konrad Lorenz researched gray geese on Lake Starnberg in the 1970s, not as much has happened in behavioral research methods as one might think. Human observations are still essential to decipher animal behavior. But these cannot capture animal behavior in all its complexity. And scientific standardization and data coupling is only possible to a limited extent.
A computer as a behavior observer
Researchers led by Joeri Bordes from the Max Planck Institute of Psychiatry have now developed a computer program that could revolutionize behavioral research. They build on the “DeepLabCut” that appeared a few years ago, with which the center points of the bodies of individual animals could already be tracked automatically. However, the program did not yet know exactly what the different postures mean for the specific behavior of an animal. However, that is now changing with “DeepOF”. For example, if you explain to the Python program that a mouse pacing frantically is expressing its stress, DeepOF can recognize this behavior and evaluate it over time.
But the program also works without prior instructions. It then identifies behavioral patterns that are similar and systematically groups them together. The scientist who looks at the identified patterns can ultimately interpret them and place them in a larger context. In this way, new behaviors can be uncovered that may have been lost in the past. Compared to previous methods, DeepOF also has the great advantage that animal behavior can now be recorded in all its complexity, even over longer periods of time or when many animals are scurrying around at the same time. Situations in which human observers reach their limits.
In addition, researchers can now combine the behavioral data in a standardized way with other measured values, such as data on neuronal or metabolic activity, and gain new insights from them. In addition, animals can now be observed in their natural setting, without the interference of human presence or an experimental setup that could distort their behavior.
DeepOF beats human observation
But what can be done with this improved behavioral observation of animals? To a certain extent, animal behavior is comparable to human behavior, especially in the context of mental illness. By finding out which symptoms a mental disorder triggers in the experimental animals and how different therapy approaches work for them, researchers can in turn develop better therapies for humans based on this. DeepOF’s first major field trial therefore took place as part of research into severe depression in mice. In order to chronically stress the animals and thereby trigger depressive symptoms in them, Bordes and his team locked one mouse with an aggressive conspecific for several months. Then they determined how the mouse reacts to other, previously unknown conspecifics – once with the established experimental setup and once with DeepOF.
The result: DeepOF was able to correctly identify the mouse as depressed and dismissive of new conspecifics. The program noticed this, among other things, by the fact that the rodent moved more slowly, looked around little and hunched over. Compared to manual observation, DeepOF was significantly more powerful and accurate in capturing the behaviors and their duration. Bordes and his colleagues therefore assume that the free open source program could also be used in observational studies all over the world in the future.
Source: Max Planck Institute for Psychiatry; Specialist article: Nature Communications, doi: 10.1038/s41467-023-40040-3