Kota Miura is the Vice-Chair of the Network of European Bioimage Analysts (NEUBIAS) and a freelance bioimage analyst. For his PhD, Kota studied cell migration in true and cellular slime molds where he was introduced to bioimage analysis, which became the focus of his research. After moving to EMBL at Heidelberg in Germany, Kota was instrumental in running the EMBL bioimages analysis courses and in founding NEUBIAS. We caught up with Kota to discuss his career, his hopes for the future of bioimage analysis and how FocalPlane can benefit the microscopy community.
What first inspired you to become a scientist?
That's a difficult question for me to answer because there was no definite step where I decided to become a scientist. When I think back, I remember attending a summer camp where my instructor was a biology teacher. He was a great guy and did lots of fun activities with us, and although I can't remember his name now, meeting him was how I became aware that being a biologist was a job. Around the same time, I started reading National Geographic and learnt about the research going on in a range of environments. Then, when I was a teenager, I did a lot of mountaineering and I thought it would be good to combine studying science with working outdoors. So, while I was completing my undergraduate degree, I started doing research on the monkeys living in the southern islands of Japan. There are many monkeys, and the research teams were studying the social behaviour of the monkeys. I attended as a helper every summer and it was hard work! You had to wake up before the monkeys and then wait until the monkeys woke up and start your observation. We followed the monkeys until they started sleeping in trees, then after confirming that they really were sleeping, we'd go back to the tent to write up our observations. It's a really tough job, and even though I was fit at the time, it was too much! For my undergraduate, I was in the faculty of science but at a liberal arts college, the International Christian University in Tokyo. I had already started doing microscopy, but even after writing my undergraduate thesis, I was still not determined to become a scientist. It was more about my curiosity for the subject, and so I thought I probably should get a PhD. I moved to graduate school in Osaka, where I started to study the cell biology of the true slime molds, Physarum. There was quite a bit of imaging, and we were quite advanced as we had a cooled-CCD camera that was only otherwise used in Astronomy back then. I was already analysing images to study cell migration; we would put tracing paper on the monitor to trace the cell area, cut it and measure the weight. That was the initial image analysis! But soon we started to use NIH image, a computational tool developed by Wayne Rasband, which later became the famous ImageJ. Doing research was fun, and I really enjoyed meeting authors of papers that I had read at scientific conferences. I also enjoyed reading the biographies of famous scientists, such as Barbara McClintock, and superstars of molecular biology. By the end of PhD, I was not questioning whether I would become a scientist, and I'd even forgotten that I was questioning this when I started my PhD!
How did you find the move from Japan to Germany, both scientifically and socially?
So, scientifically I switched from studying true slime molds to cellular slime molds. In Japan, I was studying cell migration in true slime molds, which migrate without external cues. I wanted to understand how migration is organised without a cue. I tried to address this question by myself, but I think it was way too difficult a problem for a master level student. When I moved to Germany, I started working with Dictyostelium. My advisor, Florian Siegert, suggested that I study phototaxis of cellular slime molds. It's an interesting question because the single cells are not able to respond to light, but when they get together as tens of thousands of cells, forming a structure called a slug, they start to do phototaxis, and we wanted to know why. Having the external cue meant that we could control the external stimuli, and then do different types of analysis rather than just observing behaviour, as we did in the absence of cues. Research wise, it was more enjoyable and more scientifically productive.
Then of course, the culture is different in Japan and Germany. One thing that I really liked in Germany was that people talk more openly with researchers from outside the lab. In Japan, people talk within the lab, and don't communicate much with other labs. When I was in Osaka University, the lab was relatively rich, equipped with good instruments, microscopes and all the other things, but in Germany, one of the instruments that I was using was from the 1920s! But overall, the biggest difference was the change of topic.
Do you have any advice for researchers getting started in bioimage analysis or wanting to become bioimage analysts?
I think that bioimage analysis is a foundation of biology. Back when I was undergraduate, we had to make sketches from looking down a microscope, and this, in a sense, is manual segmentation of biological structures. Now we use computers, but the knowledge that you gain is not so different except for the resolution we can achieve and the amount of data we can acquire. The essence of research is the questions we ask based on those observations and measurements of biological systems, and that essence is not changed by the advances in technology. While it is good to learn computer programming, this isn't enough to make you a good bioimage analyst; you need to know about biology and how the questions are generated, because this is directly connected to how you address the question using the tools of image analysis. Merely trying to use ‘the cutting-edge algorithm’ might not take you to the core of the question. Getting the answer by creatively bringing together biology and image analysis is very intellectually exciting, and the true originality is in how these two disciplines are combined! However, it can end up as an ‘unhappy marriage’ if the biologist doesn't understand anything about computer algorithms or the computer scientist doesn't understand biology, because it leads to a lack of constructive criticism and ultimately a potentially faulty analysis. This means that as a computer scientist, you can't be shy about asking questions and reading papers about biology and vice versa for the biologist. These discussions are essential for a successful collaboration.
What are the benefits of being a freelance bioimage analyst – are there any downsides as well?
Like becoming a scientist, this wasn't something that I set out to do, it happened because I finished a contract just as my son started going to school. I was already stressed as a single father, and when he started school, I realised that I needed to be there when he came home because he couldn't open the door – the lock was stiff and complicated. I was in Heidelberg, driving down from EMBL, rushing because I had to be on time, and I felt that it was too dangerous. This coincided with my contract ending so I decided to try freelancing. It's good because it is flexible, and now I'm living in Japan, which was prompted by the realisation, during COVID lockdowns, that I could really do all my work remotely. Going back to work at an institute is still something I would think about doing in the future though. I enjoy the flexibility but of course, as a freelancer, you have to find the people that you'll work with, and you also need to have some parallel work. It's a balance between not being too greedy and being conscious about getting the next project – this can be stressful. I also miss the style of basic science, where the answer is often not so clear. This means that you need to do some exploration and as you explore, you start to have new questions that you need to address. With freelancing, it's not like that. You'd better be clear about the task; what you'll cover and the end point. Otherwise, you can't estimate the cost and so on. This can be difficult if you are working with academics, but with companies it's much easier to do. However, in the end, I don't find this as exciting as doing something uncertain. And then you have to send out invoices, so the relationship is very business-like and doesn't feel like a collaboration. Overall, I think freelancing in image analysis can suit someone that is optimistic and mentally tough enough to write invoices!
Can you tell us a bit about NEUBIAS, the bioimage analysis consortium that you helped to launch?
Back in 2014, the biological community was not sure about the distinction between software development and image analysis. NEUBIAS started to try to convince the biological community that there are specific experts for image analysis in biology. Previously, the ‘computer guy’ was the person they went to with analysis questions, but there are people that are very good at software development and people who are good at image analysis, and this ‘computer guy’ might not have the knowledge that you require. So, the first thing that NEUBIAS wanted to do was to advertise to the biological community that there are people doing analysis, and they are not developers. We call these experts bioimage analysts and that has propagated, which was our aim. The second aim was to explain the difference between image analysis and image analysis in biology. So, image analysis in the digital image processing world, in computer science, is about how to mimic visual recognition of a human using a computer. You try to create an algorithm that mimics human recognition; this is the development of artificial intelligence (AI). But for image analysis in biology, we try to use the computer to measure things as objectively as possible and try to avoid human recognition. Now, AI does have a very important role in bioimage analysis, for example, in cancer diagnostics, where deep learning-based methods are trained using images of histological sections annotated by humans, and where we can match this to what happened to the patient – together this information could be used for automated diagnosis. But when incorporating AI in our bioimage analysis, we must be aware that some training data is based on subjective, human, decisions. We feel that it is important that biologists avoid the confusion between subjective and objective image-based measurements. The third aim, which started in 2023, is to create a global society. We received support from Chan Zuckerberg Initiative (CZI) to build this society, which we have named the Global Bioimage Analysts Society (GloBIAS). The building of GloBIAS is being led by Robert Haase and other analysts who were initially the students of NEUBIAS training schools. In many institutes, there is only one person, or maybe even nobody, who is an expert in image analysis, and we hope that the community means people feel less lonely, as well as having a network of people to share our work and exchange ideas to reduce overlaps of efforts. Just like NEUBIAS helped bioimage analysts in Europe get to know each other, we now want to network with lonely analysts worldwide!
NEUBIAS started to try to convince the biological community that there are specific experts for image analysis in biology.
As you already mentioned bioimage analysis requires collaboration – do you have any advice on how to establish a successful collaboration?
The best practice should be for the biologists to go to the bioimage analysts before they actually do their experiments. Instead, we (bioimage analysts) are often presented with one year of imaging data and asked to find something out of it. Very quickly we see many failings, for example the resolution is not consistent, or the experiments have different timings, and it is not possible to compare the data. I recommend that the biologists should invite the bioimage analysts to lab meetings if they think that a project will require image analysis, so they can get and give feedback. It's more efficient and more productive.
Bioimage analysis is a rapidly growing field, what are your hopes for the future?
A long-term goal of mine is to convince the biological community that bioimage analysis is a foundation for studying, or doing research in, biology. I strongly believe that the community should insist that this becomes a topic covered in undergraduate degrees, because it's one of the basics. In analytical chemistry, which I studied at undergraduate (and in fact nearly became an analytical chemist), you spend a year learning how to measure certain chemical species. You start with how to measure mass, and then you need spectroscopy and chromatography and so on; you learn a lot of different measurement methods. I think that image analysis should take the same role; it's about learning how to measure. And biological systems are much more complex than physics or chemistry, and images are multi-dimensional data, data in space, time, the spectrum (molecular species), and so on. How to treat those multi-dimensional data should be taught systematically during the beginning of biological studies, so that you can proceed to do good quantitative biology. What I'm trying to do with NEUBIAS, and now to globalise, is to convince the biological community to squeeze bioimage analysis into the basic curriculum in biology.
A long-term goal of mine is to convince the biological community that bioimage analysis is a foundation for studying, or doing research in, biology.
What are the current topics in bioimage analysis that you're excited about?
There is so much going on with new plugins, new tools, new platforms, it's a lot to keep up with. This includes deep learning, AI-based methods. But what is more interesting to me is seeing these tools applied in cell biology or developmental biology. I like going through these papers and checking for clever ways of doing image analysis. Of course, people get excited about new tools, but it is often the combination of different tools, even if they are old and already well-known tools, applied to great experimental design that produces the most exciting biological results by using image analysis algorithms in unique and original ways. For example, most people do not get excited about new tools in statistics – what people will become fascinated in is how statistics is used to solve certain biological questions. I’ve become more excited about how image analysis produces new biological knowledge, but less with new tools that are published on an almost daily basis.
You have spoken and written about reproducibility in image handling and analysis, why do you think this is such an important topic? What are your top tips for making sure that your analysis is reproducible?
Often reproducibility is emphasised in terms of that other people can do the same, but in the paper that I wrote with Simon Nørrelykke we tried to address the question around how to avoid data manipulation, fake data and so on. We'd been talking about this a lot and our conclusion is that we cannot really avoid it in the end. In the past 5 years, generative AI has been able to create good data and it will become more and more difficult to recognise what is real. But the important thing for reproducibility is that when you write a paper, you write the method, you write the results as objectively as possible, and it should be possible for someone to reproduce the experiment and the results can be verified. It's the same for image analysis, it should be possible for people to check if the image analysis is done correctly. This is particularly important because people often lack the knowledge of how to handle image data and might not intend to fake or manipulate their data. Only a small fraction of scientists intend to fake their data, but if we have the convention of writing reproducible methods it will be more difficult for these people to get away with it. There's a long history of science with a lot of fake data, and that it's not that image data is specifically somehow prone to faking, but we need to ensure that people write methods as clearly as possible so the data can be verified – that was the message of our paper.
How does FocalPlane and the FocalPlane Network benefit the microscopy community and in particular the bioimage analysis community?
I enjoy reading the preprint lists that you post with different categories, including cell and developmental biology and bioimage analysis. I often spot preprints that I haven't come across. I think this is very good work. The second thing that I like is that the site can be used as an information exchange platform in a relaxed way. I think it can make a big contribution to the imaging community.
Can you tell us about the process of writing your books on bioimage analysis?
We have three books now, published in 2016, 2020 and 2022. The first one was actually based on the course that we organised at EMBL in Heidelberg, starting in 2013. From the initial course, I'd proposed to the trainers that we write a manual for each module as if we were writing a textbook, and we had six or eight modules, maybe even more than that at the start. These texts really helped because the people on the course had different levels of knowledge or experience, and while some people could follow with words and explanations, others really needed to read to understand what was happening. It also meant that if people lagged behind, they could still follow the course. So, from the beginning of the course, we had very intensive texts including figures, and after the course I thought I really should make a textbook out of this resource. The first book was published in 2016. The book was really popular and this, along with the fact the EMBL courses were always over-subscribed, motivated us to continue with the same style. For the second book, NEUBIAS had started, and more people were involved. We wrote the chapters after a course, which by this time had fewer topics. We have always tried to show how we would approach biological questions using image analysis, which we call ‘bioimage analysis workflows’, and people seem to like this style. The book published in 2020 has had 111k accesses by now. And there are still many questions that we haven't yet addressed! All these textbooks were published with open access, and we still need financial support to continue in this way. We hope that building GloBIAS will contribute to the sustainability of such contributions to the biological community through eligibility for grants or calls for crowdfunding by the society.
Finally, could you tell us an interesting fact about yourself that people wouldn't know by looking at your CV?
I think it might have surprised people to know that I started my scientific career by chasing monkeys! I chased monkeys and then chased cells before moving to bioimage analysis.
Kota Miura's contact details: Bioimage Analysis & Research, BIO-Plaza 1062, Nishi-Furumatsu 2-26-22 Kita-ku, Okayama, 700-0927, Japan.
Kota Miura was interviewed by Helen Zenner, Online Editor at Journal of Cell Science. This piece has been edited and condensed with approval from the interviewee.