SUPPORTING IMAGE ANALYSIS AND COMPUTATIONAL INFRASTRUCTURE
Christian Tischer is a specialist in microscopy and image analysis. He worked for 10 years at EMBL’s Advanced Light Microscopy Facility, and since 2018 leads EMBL’s Center for Bioimage Analysis. His project is multi-pronged, including driving forward the establishment of standard file formats for imaging data, developing a unified teaching curricula to ensure world-wide accessibility to image analysis training, and exploring novel tools for the development of efficient and interoperable analysis workflows.
What was the inspiration for your project? How did your idea for the CZI project arise?
CZI funded work that I was already doing or planning to do. CZI wants to broadly support people who do bioimage analysis and microscopy.
What did the CZI enable you to do in terms of your project and goals?
Before I even applied to the CZI, I negotiated with leadership in my institute: if I get the money, because then my salary would be covered, I would like to use the money to hire someone else to work with me, because I’m not enough. That was the deal. It allowed me to ‘double’ or significantly increase the capacity of our institute- this was a very tangible outcome. It was also huge for my career – it helped me become a team leader now. The grant helped internally and politically, to promote me. This is also something CZI is hoping for: they support scientists with the hope that something happens at their institute, that significantly and sustainably changes that place. I don’t think they want to forever give money, but rather, they hope this can trigger some changes in the mid term, that it’s a catalyst. They are also acknowledging that our jobs and roles are useful and valuable. In the field of bioimage analysis, Kota Miura has done seminal work – I think the word bioimage analyst did not exist 10-15 years ago. Kota created a network of European bioimage analysts and the main vision was to make this a proper job, because before it was non-existent as an official role – it was some postdoc or PhD student helping a little bit here, and a microscopist helping a little bit there. Kota’s vision was making ‘bioimage analyst’ a job description. He was really successful, so I think Kota deserves a lot of credit here. I think CZI wants to help in this direction: if people get money from them, the recognition from their own institutes increases.
How do you handle image analysis at EMBL in your role?
We currently don’t charge anything, but I know other places do, and this is a big discussion. I have colleagues who are never offered authorship in papers with the excuse that people have to pay for their service. And if they pay, they don’t put you as a collaborator. I have other colleagues in other countries for whom this is never an issue. So, it’s a big topic without clear guidelines.
We got funding from CZI to continue networking bioimage analysts – we want to make it more global, like a bioimage analyst society. Something we want to offer is exchange of experience meetings of people who do service already, and these people can discuss how it works for them, how they charge for services, how authorships are negotiated, etc. We want to discuss what works and doesn’t. At EMBL, our work mainly supports internal scientists and visitors, which are people who visit EMBL and if they do a microscopy project, we also help them with the analysis. While visitors projects can have more tight timelines, the internal projects can last very long, sometimes a whole PhD. For those projects we therefore generally don’t have very tight deadlines, and so it’s really open – we accept every type of project. It generally works out because their experimental design also takes time: getting their gene expression or their probe or animal model – so there are natural delays. Still, I’m always afraid to overcommit and then not be able to deliver. But so far, probably because there typically no strict timelines from people who seek our help, there has been no tension in this respect.
You are bridging a major gap in expertise with your job. There are the computer scientists, and then biologists working together. What do you feel is the importance of facilitating communication between people who do microscopy, biology or chemistry and those who do image analysis?
This is a question of what people teach at universities. If you do microscopy, you should be taught image analysis too. EMBL is not really a university, so I don’t have good knowledge about what’s happening in the German education system. Some things are improving compared to the past, but it’s not like we now get PhD students who are proficient at image analysis from their prior education. I’ve actually never met someone who has had a formal education in image analysis. But that’s why we try to do this, and why teaching is a big thing for us. We try to enable people. This is something I wrote in my CZI proposal – one thing I did with some colleagues is I thought it would be great if there was bioimage analysis material that would be reusable and updated all the time. The state-of-the-art is that every trainer comes with their own PPT slides and their own way of doing things, but I had this ambition of creating something where several people could teach from the same teaching material. Something unified. We have a website about it. There are now a handful of people who actively contribute to it. We have taught about 11 courses, with 300 participants, everything is git based, and it can be updated and improved. At a CZI meeting 2 years ago, there were African scientists who said that the main issue is not having people with enough expertise. This triggered in me this idea of how training and exchange of experince should evolve. It doesn’t help if I travel to Africa to teach image analysis. In general it doesn’t help if a few people in the world travel everywhere to teach people basic Fiji. It’s better to train people in every continent and country, to do the teaching themselves. Now I’m much more interested in the “train the trainer” approach. I can think of a success case whereby I was invited to Japan to teach at a GBI course, and I initially refused to go – I asked them instead to find two people in Japan who would be willing to do the teaching. I offered to have Zoom sessions with them to show them this training material, and they would then teach the course themselves. This worked, and I was so happy about it. This also brings me to the topic of recognition of the value of our positions. The reason why some countries don’t have anyone, is probably that there is no bioimage analyst identified and recognized. Together, I feel the curated resources we have put together, and encouraging countries to recognize the job of the bioimage analyst through the train the trainer approach is something valuable and game-changing. An additional advantage of this approach is that you get people trained in their own language, further reducing barriers.
What challenges have you faced in your project, and what are your biggest successes?
I think our creation of the training resources was a big success. I am really proud of it. Also because it’s tough to get a community to support something, so this was nice. Current challenges are lack of hard-core developers. I think the tricky thing with bioimage analysis is that you’re not a developer necessarily, because you have to be good at so many things – you have to be good at biology and microscopy and so many disciplines, so you simply may not have the time to support the hard-core software development. In the whole Fiji Java ecosystem, we are lacking good Java developers who are interested in bioimage analysis. The Java ecosystem is hard to maintain, and there are crucial things missing in Fiji and it’s very hard to find people who can do this, even with CZI funding. You just can’t find anyone to hire. Probably good Java developers don’t think about bioimage analysis. Or they probably go to big companies with higher salaries, so it’s difficult to compete altogether. It’s similar with deep learning – keeping up with everything is impossible, it’s an exponential growth on a lot of information. The other big challenge we have is saving images – lacking a unified format – this is the daily nightmare. And for instance, nothing will be THE format if it can’t be opened in Fiji. It would be a show-stopper. So again, you need Java developers. And human resources are scarce – finding professional programmers who get involved in this area altogether is very rare. Maybe in the Python world it is a bit easier to find developers, but very popular software like Fiji and QuPath are written in Java. Another thing that frustrates me is that there is this gap between the open source community and commerical software. For example Imaris – it’s such a fantastic piece of software, and in 20 years we haven’t been able to produce anything comparable that is open source. They had a good file format from the start. They made all the right choices, they have beautiful 3D rendering, they were ahead of their time. In open source land we are nowhere close. Maybe CZI can run a workshop on how to bridge the gap between what companies like Imaris and Zeiss do, and what the open source people do, so there’s less of reinventing the wheel all the time.
Where do you see the state of the art for image analysis in 10 years? Do you think the issues you mentioned above will be solved?
What we are trying to do now is improve things: getting better at designing image analysis workflows because right now, for many people, they use an ImageJ macro as a workflow which is fine to some level, but it doesn’t run well in a cluster, it doesn’t scale, it’s difficult code to maintain, etc. That’s where we are very involved, in trying to use established workflow management systems like Nextflow and Galaxy for image analysis. It that works, it might change things. Zeiss has Appear, where you can do scalable, reproducible workflows. I co-organized a hackathon in Zurich, and we hope to discuss in which direction we want to go. This discussion on how we can make reproducible, scalable workflows is a big topic, and I’m optimistic that in 5 years there will be good solutions. I’m also optimistic that in 5 years we will have solved the file format issue as OME-Zarr seems to be broadly accepted. Otherwise, I think another big thing is AI – what I’m seeing is that ChatGPT is really helpful in writing code. My vision is that if researchers can read Python code, they can ask ChatGPT to write code for them to solve specific issues, and they can inspect it and determine whether it makes sense or not. This will dramatically help in terms of enabling life scientists to perform scalable and reproducible image analysis. I think this could be a game-changer.
Check out our introductory post, with links to the other interviews here