| DR. THOMAS NAERT |
| Dr Thomas Naert is a post-doctoral researcher at the Lienkamp Lab in the University of Zurich‘s Institute of Anatomy. His work recently earned him a prestigious fellowship under the Marie Skłodowska-Curie Actions programme. |
Combining his interests in the technological and the natural, Dr. Naert‘s research investigates how CRISPR and machine-learningbased tools can be used to model congenital kidney diseases in Xenopus tropicalis.
We sat down with him to ask him about his work. |
| THANK YOU FOR TAKING THE TIME TO SPEAK WITH US! I KNOW YOU ARE WORKING WITH XENOPUS AT SOEREN LIENKAMP’S LAB IN ZURICH. WHAT LED YOU TO THIS GROUP? |
| I moved to Zurich in February 2020 (just in time for the start of the pandemic!), after finishing my PhD in developmental biology at Ghent University in Belgium. I did my Masters and PhD working with frogs, so I was really looking for another Xenopus lab, ideally one that was quite tech-driven. For me personally it’s always been a bit more about the technology; that’s why I previously worked with CRISPR/Cas9, back when it was new. Soeren Lienkamp’s lab is very technologically driven. |
| WHAT IS THE CURRENT FOCUS OF YOUR RESEARCH? |
Currently I’m working on applying machine learning to disease modelling in frogs, mostly with a focus on polycystic kidney disease. We start by using CRISPR/Cas9 to create tadpole models with polycystic kidneys. That part goes very quickly; it only takes 4-5 days for the cysts to appear. Then we put some dyes into the embryos, and use the MesoSPIM open-source light-sheet microscopes to get 3D scans of the kidneys.
The MesoSPIM was developed here in Zurich by the Helmchen Group at the HIFO in Zurich. It’s like an MRI, but instead of using resonance, it uses lasers. It allows us to very quickly get 3D scans of an entire embryo. The problem is, the image files are huge – about 16GB per animal, so manual analysis would basically be impossible. That’s where deep learning comes in. |
| SO YOU GET ALL THIS RAW IMAGE DATA, AND INSTEAD OF ANALYSING IT MANUALLY, YOU USE DEEP LEARNING TO FIND PATTERNS IN IT? |
| Exactly. The machine does most of the work now. We have set up a platform by training the algorithms with the end goal to create some kind of standardized platform where we can investigate different conditions, drug tests, etc. Before CRISPR/Cas9, it used to take years of work to make animal models, but now it’s relatively simple. Making an animal model is not really an issue anymore, so now the question becomes: How do you get meaningful information from it? What we are doing now is training deep learning algorithms to automate that phenotyping process. This has the added benefit of eliminated researcher bias. |
| TRAINING SOFTWARE SOUNDS LIKE A COMPLICATED FEAT. |
| I’m a biologist by training. Until 1.5 years ago I had never written a single line of code. But the nice thing about deep learning is that, the more sophisticated the program gets, the less human intervention is needed. We still need to do some manual annotation here and there, but it doesn’t require a computer science background. Up until a few years ago, you might have needed a computer science degree just to work with these neural networks, but now I think most people can learn it with very little effort. And that will become more and more important as datasets in science keep getting bigger and bigger. |
| WHAT MAKES FROGS AN IDEAL MODEL FOR THIS KIND OF WORK? |
Personally, I really like how quickly you can do experiments. Xenopus spawns a large number of embryos, and in some cases, you can get kidney disease modelling results within weeks. And since amphibian animal models have external development, it’s easy to look at early development with frogs. Early development as always interested me, that’s one reason I used to work on childhood cancers.
And CRISPR/Cas9 has made everything even easier. If you tell me a gene to investigate, I can get it done in 2 weeks – with a mouse it could take up to a year. We think that’s a really nice baseline. Clinical genomics has changed a lot in the last 10 years, but it’s still not feasible to make mouse models for everything. Frogs and Zebrafish can be used as an intermediate model: You can look at certain genes and eventually narrow it down before moving to a higher model, like mice. We can do this very fast now, and the more automated this becomes, the more genes we can screen. |
| WHERE DO YOU SEE THIS TECHNOLOGY BEING USED IN THE FUTURE? |
| The deep learning platform we are currently setting up, and the work we are doing, is being distributed free of license. We want it to be used as widely as possible. |
| LASTLY, WHAT DO YOU LIKE TO DO WHEN YOU’RE NOT IN THE LAB? |
| I’m a post-doc, so I’m in the lab a lot! But when I’m not in the lab, I like to be outdoors; I really enjoy climbing and hiking. That must be one of the reasons I chose to come to Switzerland. You just have to drive an hour and a half from Zurich, and you’re already in the mountains! |
|
| Dr. Thomas Naert is a post-doctoral researcher at the Lienkamp Lab in the University of Zurich‘s Institute of Anatomy. |
|