Formerly known as European Learning Laboratory for the Life Sciences
Our inspiring educational experiences share the scientific discoveries of EMBL with young learners aged 10-19 years and teachers in Europe and beyond. We belong to EMBL’s Science Education and Public Engagement office.
By Hariharan Arevalagam
The European Molecular Biology Laboratory (EMBL) held their yearly Insight Lecture on the 12th of November, 2021. The senior EMBL scientist of this year’s lecture was Dr. Anna Kreshuk, the head of the Kreshuk Group at EMBL Heidelberg. Dr. Kreshuk gave an hour-long lecture on her field of research, computer vision in the area of modern microscopy. The lecture was broadcasted live via Zoom to more than 1000 young participants, mostly secondary school students, from over 20 different countries.
The lecture, titled “Cells, computers and microscopy: how can AI pave the way to scientific discovery?” was the 11th annual Insight Lecture organized by the education department of EMBL, the European Learning Laboratory for the Life Sciences (ELLS). ELLS organised their first Insight Lecture in the year 2010, and it has become a staple in their list of activities for young learners ever since. The Insight Lecture usually takes place for both on-site and online audiences. However due to the strict regulations in light of the Covid-19 pandemic, this year’s lecture was held exclusively online.
The lecture was opened with a short Q&A session from the head of ELLS, Dr. Agnes Szmolenzsky, about Dr. Kreshuk’s background and reasons for pursuing image analysis. This short exchange was especially relevant considering the main goal of ELLS, which is to bring cutting-edge research into the classroom, and in doing so inspiring students with science. For Dr. Kreshuk, inspiration came from growing up in a family of scientists. “Seeing them everyday and seeing how healthy and fulfilling life can be when you have a job that you actually like has made a mark on my decision […]”, said Dr. Kreshuk.
Before diving into the main content of the lecture, Dr. Kreshuk briefly gave an introduction to EMBL, and the research that is carried out there. She placed a special emphasis on interdisciplinarity, as science is becoming increasingly multifaceted. She embodies this herself, being a mathematician turned computer scientist who works with Machine Learning techniques to analyse biological data. “Modern biology requires the support of people of very many different specialties”, she emphasised. This idea would be one of the underlying themes of the lecture, and was also illustrated by the historical development of microscope images. These images started with scientists having to draw what they saw on paper, to computer-aided techniques that can provide terabytes worth of data, showing the tiniest details of the specimens being studied. However, the immense amount of data produced could make it difficult for scientists to find exactly what they are looking for. Therefore, some help is required.
Computers see the world much differently than we do. Unlike us, who can detect objects simply by looking at them, computers see the world in numbers. So computers need to find the number patterns that correspond to a certain aspect of the image that needs to be analysed. The example used by Dr. Kreshuk was a computer trying to detect individual cells from a microscope image. In order to do so, the computer needs to be taught to recognise certain characteristics of the image, or rules. Manually teaching a computer how to do this takes up valuable time. Wouldn’t it be better if computers can learn these rules on their own? This is the idea that led to the development of Machine Learning.
In Machine Learning, scientists input data into an algorithm called a Black Box. The data programmed into the box contain the exact features that it needs to recognise. It is then up to the computer to figure out the rules and detect more images that correspond to them. This is a lower level model of Machine Learning called “shallow learning”.
“Deep learning” involves a bigger Black Box. Instead of telling the algorithm what to look out for, it is fed with thousands of examples and it figures out by itself what features constitute a desired characteristic. In simpler terms, deep learning requires much less human intervention.
Now we have two domains – microscopy and computer vision. One generates immensely large amounts of visual data, and another that specialises in looking for patterns in that data. Putting them together results in the state-of-the-art imaging techniques that are being done by Dr. Kreshuk and her team. Using Machine Learning techniques, they are able to generate very detailed images in short times, and reconstruct images from the Terabytes of data produced by the microscopes.
It might appear that Machine Learning is the be-all and end-all solution to imaging complex structures in great detail, providing “free lunches”, as Dr. Kreshuk calls them, of fast and accurate reconstructions and speedy, high-quality 3D imaging. However, there is one significant shortcoming of this technology – the algorithms make strong assumptions from training data. These assumptions could lead to the algorithms producing images that are misleading and inaccurate. This means that as automated as the process may be, scientists will still have to be very careful with what data they feed into the algorithms.
After her lecture, questions came flooding in. For about 20 minutes, Dr. Kreshuk received more than 30 inquiries from the young audience members around the world about themes such as interdisciplinarity, applications of Machine Learning and Artificial Intelligence, the ethics of these technologies, and many more.
Computer vision is something that has completely revolutionised the field of microscopy, providing us with more insights than we have ever had access to. The level of detail that scientists are able to look at would have been unimaginable even just a few decades back.
According to Dr. Kreshuk, now is a great time to be a person with a computing background and work in science, as there are countless unanswered questions and unexplored new directions to go towards with this new power in biology research. Who knows what else computers will help us see in the near future?