Lab Rotation Interview: Marta Tataryn and Emma Kragelund Christensen

NAD fellows Emma Kragelund Christensen and Marta Tataryn on how computational neuroscience methods can provide new approaches to scientific questions.

The field of computational neuroscience is in rapid development, and data analysis and machine learning methods are of great interest to neuroscientists around the world as they can help make sense of large and complex data sets and provide new ways of approaching a scientific question.

Both Marta Tataryn (MSc in Neuroscience, University of Trieste) and Emma Kragelund Christensen (MSc in Neuroscience, University of Copenhagen) are trained as experimenters – however, they both came into the NAD PhD programme with a profound interest in computational neuroscience and data analysis methods. This led them to choose to do their first rotation at the Analysis Group with PI Associate Professor Diego Vidaurre, a part of the Center of Functionally Integrative Neuroscience (CFIN) at Aarhus University.

The Analysis Group develops computational models for brain activity across a range of data modalities; invasive and non-invasive electrophysiological data as well as neuroimaging data. They strive to characterise and understand the brain’s spontaneous activity and how this activity shapes perception and, most generally, our relation to the world. Since the way our brain works is in certain ways unique to each of us, these models are fitted at the individual level, which in turn allows the group to study how these subject-specific models relate to and predict behavioral traits and clinical variables.

The group works with datasets generated by a wide range of collaborators from all over the world.

From experiments to analysis

While both fellows share an interest in the data analysis field, their projects during the first rotation have been quite different.

Marta’s main research interests center around perception, memory and learning, and her project during the rotation focused on working with human data sets and improving her skills within coding and machine learning:

“My project was based on investigating spontaneous activity in the brain and especially how this spontaneous activity influences human tactile perception. For my investigation I used an MEG dataset, previously acquired in Dusseldorf University. It was based on a tactile discrimination task designed to be ambiguous. This means that that there were two closely related tactile stimuli in time, so it was difficult for each subject to say whether they perceived one or two stimuli. It was my first time working with a human data set and applying machine learning, which is a computational neuroscience method used to decode brain activity. During my project I decoded behavioral outcomes, so the correctness of the perception in this case, from the spontaneous brain activity.”

Marta feels like this rotation has helped her learn things on two different levels. One is the big picture: as it was Marta’s first time working with a human data set, she feels like she had some exposure to the complex questions that you can ask based on this type of data sets. The other level is the small details: Marta dramatically improved her coding skills, and she learned how to structure human data sets.

One thing that proved challenging for Marta was the lack of ownership over the experiments that data were based on:

“It was a new experience because I was trained as an experimenter, so it was my first time not doing experiments myself. Analysing that dataset was challenging for me as I am new to both cognitive science and machine learning. Without daily discussions with the experimenters to clarify the data structure and their research goals, I encountered some difficulties in understanding the context of the data,” she explains

At the same time, this provided her with valuable time to dive into the complex methods:

“I am happy I started my rotations this way because, being new to my project, I had to spend a lot of time reviewing literature and going through online tutorials to understand it better. This gave me the perfect chance to begin formulating my research question. During this rotation, I acquired a broad perspective of what could be feasible or not. It offered me the tools to move forward in the direction that I want to go. Also, working with human data sets helped me understand what types of questions I could ask,” Marta explains.

With her pre-NAD work centering on spinal cord injuries and spasticity in mouse models, the field of data analysis is also quite new to Emma. For her project in the first rotation, she has been focused on pseudotime analysis:

“Pseudotime is […] latent temporal information that you extract from a data set. Most typically […] single cell data sets, gene expression data. You try to look at the data and find a pattern that can tell you what is going on in the biological process of this particular cell. So you order the cells by the state that they are in,” Emma explains.

During her rotation, Emma has done work based on findings from a paper by Campbell and Yao, who have developed a pseudotime algorithm called Phenopath:

“The authors have developed a pseudotime algorithm for heterogeneous datasets by incorporating covariates to account for their interaction effects on the biological process. By doing this, they suggest that it may be possible to use this algorithm on different kind of data that is not single cell or gene expression data. I tried using it on structural MRI data, focusing on gray matter volumes. Conditions like Alzheimer’s disease are characterized by a large loss of gray matter volume, but gray matter loss also naturally occurs with age. I was trying to implement this algorithm to see if there was a way to find a pseudo-temporal trajectory in the data that allowed me to distinguish healthy subjects from Alzheimer’s disease subjects. The goal was to see whether it could be used as a diagnostic tool,” Emma explains.

Her preliminary results show that there is definitely something to build upon.

Building a basis for a PhD project

Overall, this first rotation has given the two fellows a unique opportunity to dive head-first into the field of data analysis and computational neuroscience, an avenue they are both going to explore further during the pre-PhD year. Both of them expect to use data analysis techniques and computational methods extensively in their PhD projects.

Emma is currently conducting her second rotation at the Danish Research Centre for Magnetic Resonance with PI Kristoffer Hougaard Madsen. She is working with resting state functional MRI data and decomposing the data using ICA (independent component analysis).

Marta is carrying out her second rotation at the Neurobiology Research Unit at Rigshospitalet in the group of Cyril Pernet:

“During my second rotation, I am focusing on developing a paradigm and implementing it to map reward circuits using fMRI within a reinforcement learning framework. I feel that this rotation is laying the groundwork for any project I might undertake during my PhD. I am in the process of defining my scientific question and trying to span between the most methods and perspectives I can, both to refine the question and to identify the best approach to study it. I am exploring, and I am really happy about what is coming out of it,” Marta concludes.

More news

Putting Your Lab on the Map

NAD Connect boosts your lab’s visibility to NAD fellows, students and potential collaborators and offers an overview of the Danish neuroscience community’s labs and scientists.