51吃瓜网

I would like to begin by thanking Mr. Harry Samuel for the Harry Samuel Arts Undergraduate Research Internship Award I received that made this project possible. The ARIA research project I pursued this summer with Professor Jean-Francois Poulin was titled: 鈥淎ssessing how disruptions in cortical activity during a critical developmental period impairs murine social behavior鈥�. The overarching goal of the project was to understand the neuronal mechanisms underlying the way early life-seizures lead to social impairments associated with autism spectrum disorder (ASD). The specific goal was to detect social impairments in experimental mice through automated tracking and behavioral classification.

Prior to starting my ARIA research project, I had been involved in the Poulin Lab as part of a COGS 401 research project, learning basic wet lab techniques and running behavioral tests on experimental mice. To analyze the behavior of the mice, however, multiple lab members must take hours of their time manually annotating specific behaviors, leading to potential bias between annotations and time inefficiency. I proposed that we implement an automated tracking system that can later be used to characterize behaviors using machine learning algorithms or simple analytics that would not require rote analysis.

Throughout the summer, my learning objectives were to create a robust and efficient analysis pipeline to enable automated tracking and extract behavioral data relevant to the project鈥檚 overarching goal.

The first step involved using SLEAP, a deep learning-based open-source software for tracking animal poses in videos. The initial step involved importing videos of interest, taking a frame and overlaying a user-defined skeleton on top of the mouse in the frame. The large diversity of labelled frames is then fed into a deep-learning algorithm which trains on the labeled frames to produce a predictor model. The predictor model is then used to run inferences on an entire video and attempt to correctly overlay a skeleton on top of each frame to track the mice.

Using SLEAP came with a few challenges. The model struggled with certain poses such as hunching or rearing, and with inconsistencies in video angles across different recording days. To overcome these issues, I made sure to increase the diversity of labels to include difficult mouse poses and different recording days, until the algorithm could provide somewhat accurate tracking of every video.

My second objective was to take the inferred tracking from SLEAP and somehow extract relevant behavioral data. Each video I had used had two phases, a phase with just the experimental mouse, and a phase with the experimental mouse and another control mouse, to observe social behaviors. The tracking from SLEAP not only had missing frames and frames with extra, non-existent skeletons, but during multi-animal tracking, the skeletons would switch between the control and treatment mouse, which was a significant issue, as we needed to correctly follow the ID of each animal.

To resolve these issues, I wrote a Python script to fill missing frames, delete extra instances, and manually fix each video鈥檚 track, as well as linearly interpolate between frames if they were added and normalizing the scale of each video to allow homogenous analysis of any video regardless of the day the video was recorded. Using these clean, preprocessed, and normalized tracking datasets, I calculated the distance travelled by the treatment mouse, the amount of time it spent on the periphery of the cage, estimated the number and time of close interactions between the mice, and created heatmap visualizations. I then attempted to create a machine-learning behavioral classifier for specific behaviors. This process solidified my ability to code in Python, solve problems and work independently, and deepened my understanding of how to utilize data to its fullest potential. The data and visualizations that I extracted and created will be presented in a manuscript my graduate student supervisor will submit this August.

The ARIA research project will undoubtedly shape my career and career path to pursue a PhD in neuroscience. This summer research project led me to understand my true passion and excitement for data analysis involving neuroscience data, whether it be with behavioral data or human brain imaging data. The project also facilitated my boost of confidence in coding and research in general and created a momentum that I hope to take advantage of.

Once again, I would like to sincerely thank Mr. Harry Samuel for the Harry Samuel Arts Undergraduate Research Internship Award I received that enabled me to undertake this amazing project. The support allowed me to engage fully in neuroscience research that will shape my career path while easing the financial pressures I face as an international student.