- Статьи
- Science and technology
- Mouse vision: neural network will significantly accelerate the creation of drugs for the brain
Mouse vision: neural network will significantly accelerate the creation of drugs for the brain
Scientists have developed an artificial intelligence-based system that, in a matter of minutes, makes it possible to evaluate the prospects of drugs for depression, dementia and other affective disorders for further research. For the first time, a neural network has been trained to automatically recognize and analyze the behavior of rats with genetic brain disorders, which makes the study of such diseases faster and more accurate than manual observation. The technology allows you to avoid watching videos for hours: the analysis is carried out automatically, and the data obtained is more objective and reproducible, which is especially important when testing new drugs. The drug developers told Izvestia about the prospects of such software for creating new drugs, but its functionality is limited to the study of animal behavior.
Rats under neural network surveillance
Specialists from Sirius University of Science and Technology have trained a neural network to automatically recognize and analyze the behavior of rats with genetic brain disorders. Such animals simulate Parkinson's disease, depression and other disorders.
To teach the neural network to understand animal behavior, the researchers manually marked up more than 7,000 frames, which marked key points on the rat's body: the tip of the nose, paws, and body. From these points, the algorithm has learned to independently recognize postures and determine exactly what the animal is doing: running, freezing in place, standing on its hind legs or exploring the territory. Previously, specialists had to manually analyze video recordings of animal behavior and record every movement. This approach took a lot of time and inevitably introduced an element of subjectivity into the results of the work.
To test the developed system, the scientists selected two genetic lines of rats with well-studied behavior. The first line is animals with impaired dopamine reuptake, they are hyperactive and serve as a model for disorders associated with dopamine system dysfunction, including Parkinson's disease. The second group is serotonin—deficient rats, whose exploratory behavior and general locomotion decrease. Such animals help to study depression and other affective disorders.
The automated analysis allowed us to determine exactly which behavior parameters differ in healthy and mutant animals. For example, in rats with tyrosine hydroxylase type 2 gene knockout, the system recorded a decrease in vertical activity. They were much less likely to stand on their hind legs. At the same time, overall mobility remained normal. This clarifies the understanding of exactly what role serotonin plays in regulating behavior, and shows that its deficiency affects very specific forms of animal activity, the scientists said.
Using machine learning, experts have identified which behavioral traits are most important for distinguishing healthy animals from mutants. For hyperactive rats, the main marker turned out to be the length of the path traveled, and for animals with serotonin deficiency, it was the hind legs. The classification accuracy reached 84% in the first case and 98% in the second. This means that based on the behavior of an animal, a neural network can determine with a high degree of probability which mutation it has. The developed approach is already available for use by other scientific groups.
— We have demonstrated the potential of using machine learning methods to analyze animal behavior. We believe that the developed system, combined with the proposed behavioral tests, can significantly accelerate and standardize both the analysis of behavior of various genetic lines and the screening of new potential pharmacological drugs," said Danil, the first author of the study, a junior researcher, a graduate student in the department of Neurobiology at the Scientific Center for Genetics and Life Sciences at Sirius University. The bulbs.
A system for analyzing animal behavior in drug development
The method is really capable of speeding up the preclinical stages of development, Alksander Zakharov, director of the Research Institute of Neuroscience at the SamSMU of the Ministry of Health of the Russian Federation, market expert at NTI Healthnet, told Izvestia. Automatic video analysis reduces data processing time from hours to minutes, allowing faster evaluation of the effectiveness of potential drugs in animal models. Objective and standardized metrics increase the reliability of screening, which reduces the risk of false results.
"Nevertheless, the acceleration will mainly affect the behavioral validation stage — the overall timing of drug launch depends on other complex stages (toxicology, pharmacokinetics, clinical trials), where this tool is not directly used," he noted.
From a scientific point of view, this is certainly an important methodological achievement for studying animal behavior in an experiment. Such approaches have been actively developing in recent years and are already widely used in neuroscience, added Yulia Komleva, an expert at the Foundation for the Development of Russian Science, Technology and Medicine (FRONTMED) and the NTI Neuronet market.
— The use of neural networks makes it possible to increase the accuracy and reproducibility of analysis, as well as reduce the influence of the human factor, which is really considered a significant advantage. It is also important to understand that the interpretation of behavioral patterns in animals is always limited. For example, parameters such as vertical activity or general mobility are indirect markers that cannot directly reflect complex conditions such as depression or cognitive impairment in humans. To analyze animal behavior, you should use a whole battery of tests with extensive analysis and integration of the data obtained," she told Izvestia.
The research was carried out with the support of the Sirius State program for scientific and technological development of the federal territory. Specialists from St. Petersburg State University, the Institute of Experimental Medicine, and the Pavlov Institute of Physiology of the Russian Academy of Sciences also participated in the work. The results are published in the prestigious journal of Neuroscience.
Переведено сервисом «Яндекс Переводчик»
