A new study from the Indian Institute of Science Education and Research (IISER) Thiruvananthapuram, uses machine learning tools to understand anxiety in zebrafish. The researchers are training computers to pick up subtle cues in their behaviour to identify when a fish is anxious.
Zebrafish are small, striped fish that have become quite popular in scientific research. They share a surprising number of genetic similarities with humans, which means discoveries made about them can often be translated to people. Zebrafish are also easy to care for and reproduce quickly, making them ideal for studies that require many subjects.
Zebrafish behaviours can tell us a lot about their internal state. When they’re anxious, they tend to behave differently—just like humans might pace or fidget, albeit in subtle ways. Researchers look for signs like how much time zebrafish spend at the bottom of their tank, how they move, and how often they stay still. These behaviours give clues about their anxiety levels.
In the past, scientists would watch videos of zebrafish and note their behaviours manually. While effective, this method is time-consuming and prone to errors. Human observers tend to make mistakes, and their assessments can differ from each other. After all, interpreting behaviour can be quite subjective, not to mention the hours of footage one will have to parse through to identify any patterns.
Enter DeepLabCut and ZebraLab—two tracking programs that helped track the zebrafish behaviour. These programs automate the process, tracking and detailing the fish’s movements minutely. Using data from these programs, the researchers trained a number of machine-learning models, including Logistic Regression, Decision Tree, K-Nearest Neighbours (KNN), Random Forests, Naive Bayes Classifiers, and Support Vector Machines (SVMs).
Machine learning (ML) is a type of artificial intelligence that allows a computer to use data to learn and improve. In this study, researchers used ML to analyze the data from the zebrafish. Using video recordings of the fish, the researchers fed information into the ML algorithms about behaviours such as total time spent being inactive, time at the bottom vs. the top of the tank, and their swimming angles.
The study found that certain models, such as Decision Tree and Random Forests, performed excellently in differentiating between anxious and non-anxious behaviour and could successfully identify differences in behaviour that even the trained eyes of human experts might miss.
While machine learning offers tremendous potential, it does have limitations. For one, the technology depends heavily on the quality of the data. If the video recordings are poor or the data isn’t annotated accurately, the machine learning models might reach incorrect conclusions. Future research could expand this study by integrating more sophisticated sensors that capture additional physiological data from zebrafish for more accurate predictions.
By understanding how zebrafish display anxiety and using machines to read these signs accurately, researchers can create better animal models for human conditions without the need for invasive procedures. This opens the door to developing new treatments for anxiety-related disorders not just for humans but across different species.
This research news was partly generated using artificial intelligence and edited by an editor at Research Matters