Classic characterization of neural activity may obscure short-timescale changes in neural signals. To address that, I developed a novel computational approach that combines signal processing and machine learning techniques to capture transient dynamics in the brain. By applying my method to our existing datasets, we found that a subset of neurons in the hippocampus, which have traditionally been ignored in analyses, play an important role in discriminating goals during goal-directed behavior. Further, this goal discrimination role of these cells failed in a mouse model of Alzheimer’s disease. Our approach and these results provide new methods to investigate neural mechanisms in cognition, memory, and their alteration in brain diseases.
Our team was made up of members from Dr. Annabelle C. Singer’s lab, studying the neural mechanisms of memory and how they fail in Alzheimer’s disease (AD). I developed a novel computational method to access the fast dynamics of brain signals and successfully tested this approach in public data sets [1]. Dr. Singer proposed that my methods could reveal something new in datasets collected by Stephanie, in addition to her findings about the neural connection deficits in AD mice [2]. Leaning on my methods and her collected data, we hypothesized a subset of hippocampal neurons that have mostly been ignored in the literature, play a unique role in goal-directed navigation. Thus, I planned to investigate the roles of those cells with two different analyses. I analyzed the activity of single cells, and a 3rd member of the team, Abigail, joined to analyze population-level neural activity. Abigail joined the team because she was learning this analysis method, while her own experiments were paused during the covid-19 pandemic. Our analyses showed that those cells improved goal discrimination in this task, and such a role failed in the mice model of AD [3]. We worked efficiently and enjoyed the whole process of our discovery.
The goal of this project was to develop and apply novel computational approaches to investigate the temporal organizations of brain in cognition, memory and their alterations in aging and brain diseases. By reusing public datasets, I developed the initial analysis tools in 2018. The initial data collection was completed by Stephanie in 2019. I started to reuse her data in 2020. Abigail joined the analysis in 2020, and all analysis work was accomplished in 2021. We believe the following points contributed to our success.
First, we established a collaborative environment between experimenter and data miner in a data reuse project. Though a well-documented description of the dataset by the experimenter is always helpful, it is necessary for the data miner to contact the experimenter directly to clarify details about the experiment. We regularly reported our progress to the experimenter, which was beneficial as the experimenter’s intuition about the dataset helped us to detect potential problems.
Second, we had good balance between joint and independent work. With the open and collaborative culture in Singer lab, both the raw data and the code to process the data were shared on our lab server. While we all have access to the lab resources, we saved additional versions in separated folders after modification results, to prevent overwriting the work of each other. We used Github for version control of project code. Thus, we reduced the impact of potential errors caused in a specific part of work.
Third, teamwork is important. Most of the analysis work was done during the covid-19 pandemic. We had one-on-one online meetings regularly for latest progresses, issues, trouble-shooting and future directions. In general, working as a team helped us to quickly find reach a solution for any issues.
Based above principles, we overcame the vocabulary barrier between experimenter and data miner, which is a common obstacle in reuse of data. We developed novel methods investigating temporal organization of the brain and generated new insights from existing data sets. I have been invited as a speaker to share our results at the upcoming Gordon Research Seminar for Neurobiology of Cognition, 2022. Being inspired by the project, I organized and will host a symposium titled ‘Dynamic Communication Between Regions’ to discuss the current progress of the field on studying the temporal organization of the brain in the upcoming conference of the Society for Neuroscience (SFN 2022).
While there is a fast-growing publication in science, there is also a replication crisis, where results of many scientific studies are not reproducible by other teams. For experimental work, the best solution to share all experimental and methodological details in the paper, as well as the collecting data after the initial publication if there is no conflicts. For computational studies, it is important for scientist to share details of the algorithm as well as the code. We believe sharing data and methods is important for increasing scientific reproducibility, as well as the impact of good methods. To achieve this, I shared the MATLAB code of my computational methods on Github and provided example and data from a public dataset (Resource 1). I also shared my protocol to quantify synchrony between two brain regions (Resource 2) based on requirement from a reader of my publication [1]. Our team shared our data and code on Github to reproduce the results in our publications (Resource 3 and 4).
Our team enjoyed the following benefits we got from our data reuse project.
First, it is helpful to establish scientific collaborations or strengthen existing collaborations. This is important because modern biological and biomedical research relies on teamwork. Establishing collaborations helps researchers to have more opportunities to communicate and brainstorm.
Second, reusing existing datasets can facilitate scientific discoveries. It can take months and even years for data collection in biological and biomedical research. If there are shared datasets that are suitable for testing a new idea or hypothesis, both financial and human costs can be saved.
Third, it benefits both the experimenter and data reuser. While the data reuser can save time and cost, the experimenter will also get credit for potential additional publications, which is good for the scientific careers of both sides.
Lastly, with the many benefits mentioned above, the experimenter will be more motivated to collect and control the quality of their datasets. High-quality data usually attracts more attention from other data users, which in turn generates more novel insights from the data.
