Data consists of much more than just the information on the surface. With statistics, deeper cause and effect relationships can be brought to light. This is exactly what Alexander Marx is researching as a Fellow at the ETH AI Center using artificial intelligence. One of his goals is to make predictions about diabetes in children.
In diabetics, hypoglycaemia does not usually occur randomly, such that the stock market price does not crash without reason. This means that both are also approximate, at least theoretically. In practice, however, such predictions have succeeded only in the rarest of cases. But if Alexander Marx’s project is successful, that will change for children with type 1 diabetes.
“We are working on predictive models that can detect if there is a risk of nocturnal hypoglycaemia,” explains an ETH AI Center Fellow. Falling asleep can fall below a critical threshold. With a reliable forecasting model, this risk can be avoided.”
Unraveling Cause and Effect Networks
Marx explores this hypothesis as part of Julia Vogt’s Medical Data Science group. “I come from a more theoretical background and have worked mostly with artificially generated data. AI Center aims to bring theory and practice together, which I find exciting. I am now able to combine my theoretical concepts with real data. Have to work.”
Marx achieved his academic credentials in the Saarland University in Saarbrücken, Germany. After completing his master’s degree in bioinformatics, he remained there to write his doctoral thesis at the Max Planck Institute for Informatics. His thesis examined causal discovery – statistical methods used to construct causal graphs from observational data, which make cause-and-effect networks visible.
getting predictions from correlations
One way of applying these methods is to use survey data to identify all factors suspected of influencing a particular variable. A common example would be how a person’s income depends on his or her age, place of residence, gender, education, marital status, or number of children.
Based on the correlations found, predictions can be made for individuals who were not surveyed. Marx explains that to do this, it is not even necessary to define the entire dependency chain; It is enough to derive the smallest set of factors needed to make a prediction.
From synthetic data to clinical reality
Using artificial intelligence based on simulated data, Marx used these methods to study how the activities of about 500 selected genes in a human cell were related. Ideally, in the future these methods can be extended to include all 25,000 or so genes in a cell.
Such computer analysis of gene networks would easily and quickly provide biological and medical research with a comprehensive understanding of the processes in a cell. Achieving this through laboratory experiments would require enormous effort, as scientists would have to turn off each gene individually using genetic engineering tools and then measure how it affects the activity of all other genes. Is.
The projects for which Marx is working at the AI Center need to take causal discovery methods to a new level of complexity. Instead of using complete observational data sets or synthetic data with gene expression, he now works with accurate data from clinical practice.
This makes the task apparently more difficult, as he soon found out: “Indeed, individual pieces of information, measurements, or entire data sets are often missing, and how the data is collected.” , it always varies from hospital to hospital and sometimes from doctor to doctor.”
eliminate irrelevant correlations
In collaboration with clinicians at the University Children’s Hospital Basel (UKBB), Marx’s clinical data for his prediction model included time series of pulse rate and blood glucose levels and information on physical activity, caloric intake, insulin injections and sleep quality. Is. Then it is a matter of filtering the data to exclude any correlations from the model that are not related to the research question.
If the prognosis is strong and understandable to a treating physician, the number of factors should be kept as low as possible. It is too early to predict whether the model will be successful in practice: “With our project, we are entering areas that we have not yet mastered with available methods.”
Nature, mountains and climbing in the community
Anyway, the young researcher has made a successful start in Zurich. “When I first came here in autumn, I quickly felt like home. The city is spectacular, and the mountains are really close,” he says. Marx is a passionate rock climber who enjoys living so close to nature Love it, especially the mountains.
“Climbing allows me to switch off and focus entirely on the grips. There’s also the climbing community — I love working with other people.” In Saarbrücken, he was far from the mountains and therefore mainly stoned indoors. Now that he lives in Zurich, he is looking forward to being able to visit the alpine areas more often.
exceptionally international and interdisciplinary
Marx likes the AI Center as much as he enjoys the city of Zurich and its surroundings: “The center is extraordinarily international. There is also a great variety of subject areas. Various scientists as a normal part of daily routine Having peer-to-peer discussions with key executives across topics is impactful and inspiring.”
However, the interdisciplinary character of the AI Center is not limited to social interaction. Along with bioinformatician Julia Vogt, Marx has a co-mentor in Peter Buhlmann, who specializes in high-dimensional statistics. These can be used to examine data sets that have multiple attributes associated with each object. This includes statistics on diabetes that Marks analyzes.
In addition, there is also an established collaboration with the Biomedical Informatics Group, led by Gunnar Rasch, which conducts research at the interface of machine learning and bioinformatics.
learning from different data sources
Marx himself is active in many subject areas. He has another project in which he has done what is known as Multimodal Learning. Here, the goal is to find similarities in data from different sources. For example, he combines the results of positron emission tomography (PET), which produces a 3D visualization of anomalies in tissue metabolism, with the effects of X-ray-based computed tomography (CT), which shows the layers of tissue. reveals density anomalies.
A combination of analysis of the two imaging methods, automated by machine learning, could lead to significant advances in tumor diagnosis. Vision is an AI system that finds similarities in two data sets and applies them to issue reliable diagnoses and forecasts.
First Experience as a Lecturer
For the time being, Marks is looking forward to his first lecture course, which he will deliver with colleagues from the AI Center this upcoming summer semester. “I have always enjoyed working with students, and the master’s students at ETH are at a very advanced level. The discussions always generate input that I had not considered myself,” says Marx. His fellowship thus allows him to hone his scientific skills and gain initial experience as a lecturer.
Marx does not want to make any concrete predictions about his own future yet, saying: “After my first experiences here, I believe my time at ETH has given me excellent opportunities for career options – both academia and industry. Will prepare.”
Source: ETH Zurich