Why do bees dance after collecting honey? Why do cats purr when they are touched by their owners? Dogs also shed tears? How do you know what's going on in an animal's head? Scientists have never stopped exploring the mysteries of animal behavior to study the regulation of behavior by brain nerves and behavioral evaluation in drug discovery.
In recent years, although new technologies such as optogenetics, high-throughput neural electrodes, and in vivo microscopy have made breakthrough changes in the field of neuroscience, they are still relatively simplified in terms of behavioral observation methods and collection methods.
On May 13, the latest research results of Wang Liping and Wei Pengfei's team from the Institute of Brain Cognition and Brain Diseases, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, were published in Nature-Communications. The research team independently developed a multi-perspective, three-dimensional automated behavior acquisition device, and proposed a three-dimensional behavior map framework for general animals ,"Behavior Atlas". Wang Liping, researcher of Shenzhen Advanced Institute, and Wei Pengfei, associate researcher of The Paper, are the co-corresponding authors of the paper, Huang Kang, a doctoral student, and Han Yaning, a master's student, are the co-first authors of the paper, and Professor Liu Siyuan of Pennsylvania State University is the co-author of the paper.
Schematic diagram of behavior Atlas
More accurate and comprehensive animal behavior data were obtained using multi-perspective, three-dimensional animal body reconstruction technology, combined with automated behavior analysis, which was used to identify abnormal behavior in mice with autism model.
It is understood that the behavior collection device independently developed by the research team for two years can obtain the three-dimensional movement posture of animals, and propose a hierarchical animal behavior decomposition model according to the hierarchy of animal behavior similar to language, simplifying continuous and complex behavior into action modules that can be understood by people. The study has conducted behavioral identification in mice with an autism model, and successfully achieved automatic and accurate identification of their characteristic behavioral abnormalities at the subsecond level.
AI blessing Multi-perspective three-dimensional display of animal behavior posture
In the traditional collection and analysis of animal behavior, a single perspective and 2-3 physical feature points are used to photograph and track animal behavior. However, the occlusion of body parts can easily cause deviations in the viewing angle, resulting in behavioral data analysis errors. In addition, there are too few tracking points to accurately collect fine limb-like movements.
All along, the team of Wang Liping and Wei Pengfei has been committed to using optogenetics technology to explore the regulation of behavior by brain neural circuits. Since 2018, the team has been thinking about using new machine learning and other technologies to solve the current bottlenecks in the field of animal behavior collection and analysis.
"Early in the study, when we used traditional methods to observe behavior, we found that the existing dataset of animal behavior experiments was lacking, and manual labeling of raw behavior videos required a lot of work, time and effort. Even if enough datasets are built for a particular species in a specific scenario, the accumulated datasets cannot be reused due to the differences between different species and behavioral scenarios. Wei Pengfei said.
In this regard, combined with artificial intelligence technology, the team independently developed a multi-perspective, three-dimensional automated behavior collection device, which can build a three-dimensional, multi-body parts of the animal's movement skeleton, and more comprehensively characterize the animal's movement.
"Compared with other three-dimensional animal behavior acquisition devices, the number of cameras used in our system has relatively low requirements for camera performance, which greatly facilitates the acquisition of three-dimensional behavior data of animals, and is very easy to expand to large animal research in dogs and non-human primates." Wei Pengfei introduced.
Previously, the research group of doctoral students Liu Nan and Han Yaning have successfully made initial progress in the analysis of anxious behavior in animals using the system, and the relevant results were published in the Biochemistry and Biophysical Research Newsletter. It is understood that the team has received cooperation and procurement intentions from more than 20 scientific research institutions at home and abroad. The two patents involved in the technology have been authorized for transformation, and the equipment has been put into mass production.
Schematic diagram of behavioral acquisition equipment and equipment put into mass production.
It took two years to build a "dictionary" of behaviors
In the process of research, exploring the essential laws of behavior, decoding behavior, and building a "dictionary" of behavior have become the focus of the team's research.
The team found that when the animals showed their behavior, they first showed a rich posture through their limbs, and the continuous change of posture formed one by one walking, jumping, climbing, scratching and other actions, further expressing the animal's nature, habits, moods and sorrows and other behaviors. This "gesture-action-behavior spectrum" model of behavior is like the "letter-word-statement" pattern in language, which is hierarchical. Therefore, behavioral data also needs to be collected hierarchically.
To accurately collect behavioral data, you need to go back to the behavior itself and build a "dictionary" of behavior. Combined with artificial intelligence algorithm technology, the research team first discretized the representation of the gesture layer of the behavior from the continuous three-dimensional behavior sequence, so that the machine can identify each gesture of the behavior, which is equivalent to building the "alphabet" of the behavior; secondly, from these gesture representations, the gesture sequence with similar arrangement pattern is excavated as the action layer of the behavior, which is equivalent to the "dictionary" of the construction behavior.
"Just like observing a mouse that is scratching, we need to deconstruct how many times the mouse scratched, the length of the scratch, the scene it is in, and collect behavioral data." Wei Pengfei said that according to the hierarchical performance of behavior, the research team further constructed a hierarchical animal behavior map. It is reported that in the verification experiments conducted by the team on the mouse model of autism, the specific behavior of the mouse model of autism was successfully found from more than 40 behavioral subtypes isolated. After analyzing the behavioral data, the behavioral results are highly consistent with the genotype of the animals. In addition, using the new characteristics of anxious behavior extracted, the team also succeeded in isolating anxiety from normal animals that could not be effectively separated by the traditional anxiety measurement paradigm.
At present, the behavior map is still being updated and improved, and it is developed for different species. It is understood that the team has carried out research on non-human primates and dogs, and will be applied to experimental animals such as pigs in the future.
Let behavioral data answer biological questions
Behavioral observation and quantification are of great significance for drug development and disease diagnosis. Take the classic "forced swimming experiment" as an example, by observing mouse behavior, it is used to detect the effects of antidepressants. However, in this process, the obtained behavioral data were inaccurate, resulting in differences in the methods of establishing and evaluating depression models, making the experiment controversial.
With the development of machine learning and artificial intelligence technology, "computational neurobehaviorology" aimed at achieving high-precision and automatic quantitative behavior is becoming an important emerging research field, and China has rarely been involved in this field, and it is still in a state of "running" internationally. In the field of computational neurobehaviorals, this study provides an effective means for exploring neural activities such as brain regulation of behavior, and further assists in the diagnosis of diseases and the evaluation of drug efficacy.
Duan Shumin, an academician of the Chinese Academy of Sciences and a professor at Zhejiang University, said that the three-dimensional animal high-precision behavior analysis system independently developed by the team partially fills the shortcomings of existing behavioral analysis. The system will greatly broaden the paradigm of animal behavior research, improve the accuracy and efficiency of behavioral analysis, and its application is expected to provide an important means for studying the onset mechanism of mental illness and mood disorders, as well as preclinical research and development tests of various therapeutic methods.
"For a long time, due to the limitations of technology, our observation of animal behavior has remained stuck in the 'where' and 'how fast' animals are running, without knowing what animals are doing. Based on the refined animal behavior detection and analysis system developed by our team, we can more objectively, accurately and comprehensively measure the characteristics and correspondence of animal behavior changes under the control of specific neural circuits, drugs and gene manipulations, so as to further promote the research of brain cognitive mechanisms, brain-like intelligence and the research and development of new drugs for brain diseases. Researcher Wang Liping, co-corresponding author of the paper, said.
Next, the research team plans to combine Behavior Atlas with miniaturized in vivo two-photon fluorescence microscopy, high-throughput neuroelectron electrophysiology and other techniques to study the functional networks of neurons and how various neurotransmitters in the brain finely encode specific behaviors in the context of in vivo, long-term observations.
Screenshot of the paper online
Source: Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences