Veronika Koren is a computational neuroscientist at the University Medical Center Hamburg-Eppendorf (UKE) in Germany, who wants to make mathematical models of neural networks as realistic as possible. In this interview, she discusses how her recent eLife article was an exercise in both personal resilience – she was unemployed when she started the work that led to the article – and scientific resilience. She also discusses how the project took shape, what kept her going, and why bridging the gap between theory and biology remains, for her, an irresistible pursuit.
What was your recent eLife article about?
Veronika Koren
Most neurons in the brain communicate using brief electrical pulses called spikes. The efficient coding theory is the idea that the brain encodes sensory information with these spikes in the most efficient way possible. Current mathematical models are elegant but do not take into account many factors in a real biological system. My research involves extending efficient coding theories of neural networks to make them more realistic from a biological point of view.
To make the models more realistic, we – Simone Malerba, Tilo Schwalger, Stefano Panzeri and myself – added features found in real neural circuits, like excitatory and inhibitory neuron types, structured connectivity, metabolic constraints, and natural background noise. The paper brings together several results showing that, even when you add realistic features, we can still achieve efficient coding, and many of the solutions of our model closely resemble the patterns seen in real systems.
How did your eLife article come about?
I completed my PhD in computational neuroscience at the Technical University of Berlin (TUB) just before the COVID-19 pandemic started, and I was looking for a postdoc position, but most labs had stopped hiring, and my husband and I were reluctant to move for safety reasons. When writing my thesis, I revisited a past project of mine on efficient coding, which renewed my interest in this topic, and I had so many ideas that I wanted to pursue, so I started working on it again. In fact, I developed most of the analytical results in the eLife article while working at home at the beginning of the pandemic, and with no salary. In June 2020, thanks to financial support from the Berlin Equal Opportunities Program, I was able to go back to TUB, where I started working with Tilo Schwalger. In 2021, however, conscious of the need to change institutions for the benefit of my CV, I started searching for a new postdoc position that would allow me to continue my work on efficient coding. I eventually joined Stefano Panzeri’s lab at UKE in February 2022, and, at first, my project stayed on hold while other work took priority. Fortunately, however, Stefano became interested in my results, appreciating the analytical development, and he helped me revise and shape them into a paper.
What made this particular question so compelling to you that you kept working on it even without funding or a lab?
My intuition told me that extending a math-based modelling approach to a network design that is closer to biology is interesting from both a methodological and a biological point of view. Indeed, as I put myself to work, this extension came out very naturally, which is why I felt really enthusiastic about it. Lots of theoretical neuroscience is about simple models, and models will always be approximations of biological neural networks. However, incorporating aspects of biological complexity can help us find better solutions to computational problems.
The first version of the article was described as "Incomplete" in the eLife Assessment. How did you feel about this, and what changes did you make to take that assessment to “Convincing”?
Some important questions about brain biology simply don’t have straightforward answers. Our paper does not have a simple storyline but rather draws a somewhat intricate but deep link between a mathematical theory and the functioning of biological neural networks. We were worried that many journals would prefer a simpler "story", but estimated that eLife might be one that simultaneously enjoys an excellent reputation and where our results might be appreciated.
I was disappointed by the first assessment, because we omitted more detailed analyses to make the paper more focused. The interest of the work comes from the ensemble of several results, which is hard to turn into a simple narrative. We tried nevertheless, and perhaps we tried too hard because, surprisingly, the thorough reviews asked us to dig deeper, which we gladly did. I followed every comment of each reviewer with great attention, significantly reshaping the paper and making it more solid.
You talked about the pressure to relocate for your career. How do you feel now, having stayed in Germany?
The environments I have worked in have given me a variety of experiences. My PhD lab emphasised methodological rigour, my first postdoc emphasised analytical approaches, and now, I focus on the biological relevance of models. European neuroscientists are typically advised to go to the US for their postdoc to open doors for a faculty position back in Europe. I find Germany surprisingly conservative, tending to reward researchers who have already been recognised elsewhere, rather than actively supporting independent and innovative work. Requiring relocation puts young researchers under unnecessary stress trying to combine life and work, which may discourage many talented people from academia. I find this difficult to justify, as Europe offers a rich research network that is more than sufficient to provide strong scientific training.
How do you see the relationship between theoretical and experimental neuroscience today? Being a computational researcher was an advantage during lockdown, but more broadly, does it feel like an advantage or a disadvantage?
Lockdown was certainly a more productive time for theoretical and computational neuroscientists. Computational scientists believe it is easier for experimentalists to publish in the so-called high-level journals, and historically, theorists have felt marginalised. Now, I think journals are increasingly open to computational work, driven in part by the rapid growth of artificial intelligence and its impact on everyday life, but this also brings intense competition for papers, grants and jobs. Additionally, this shift does not necessarily contribute to understanding biological neural systems, as the research into artificial intelligence has increasingly diverged to focus on technological applications.
Ultimately, though, science is about understanding natural phenomena, and I hope that discovery, whether or not it leads to technological advancement, holds its place in society. I see the future in closer collaboration between computational and experimental neuroscience.
Interview by Daisy Veysey.
About Veronika Koren
Veronika is a senior postdoc in the Institute for Neural Information Processing at UKE in Hamburg, Germany. She has a Master’s in cognitive science from the École Normale Supérieure in Paris, and a PhD in computational neuroscience from TUB.
