AI tools to predict risk of common diseases using genetic data

Åsa Johanssson and Hadrien Gourlé collaborate in a project where they study how genetic data can be used to estimate the risk for common diseases. Their specific goal is to develop a methodology that can enable the identification of individuals with high risk of developing a disease.

Åsa Johansson, PI of the project, leads a research team focusing on genetic epidemiology, that is to say how genetic factors influence the development of diseases in populations. She explains that in contrast to clinical genetics, in which single mutations that cause severe conditions are detected and studied, it is much more difficult to find a link between the genetic background and common diseases such as cardiovascular diseases or asthma.

“These diseases are to a large extent also caused by genetic variants, but instead of one mutation in one gene, hundreds, or even thousands of genes and genetic variants, are involved. To be able to link this genetic variation to disease risk it’s necessary to analyse the genetic information from a very large group of people,” she says.

I recent years, the DNA sequence from individuals in such large groups have become available. Åsa’s group is using data sets from 500 000 people that contain both genetic information and information about diseases. This makes it possible to measure the very small effects that together contribute to the disease. If all these small effects are combined it could be used to explain the risk for a disease.

Photo of a stethoscope and an image of DNA

To be able to extract more of the information in this large data set Åsa wanted to use the advanced computer technology called machine learning, which is a tool in the field of artificial intelligence (AI).

“In my group we have a lot of experience in genetics,” she says. “We are close to the data, so to say, but further away from machine learning. So, we needed someone who is proficient in the machine learning methodology and this was how our collaboration started.”

Expertise and new knowledge

Åsa and Hadrien had just started discussing the possibility to recruit Hadrien to Åsa’s group when U-Share’s call for proposals opened. And after working together for a few months now, they agree that it has been a very good match.

“From my PhD studies I knew machine learning very well but I have always been interested in research in human medicine. So, I am very pleased with joining Åsa’s group,” Hadrien says.

Åsa adds: “Having Hadrien and his expertise in the group has been very positive. At his previous department at SLU, computer sciences and genetic research seem to be more mixed as compared to the medical faculty, which is more focused on clinical applications and has fewer interactions with the computer sciences.”

Hadrien thinks that taking machine learning into a new field has so far not been extremely difficult since the technology is rather similar wherever it’s applied. But he has been forced to learn a lot of new things, specially about the diseases and the human genome.

“During my education I have taken several courses in human genetics but it was more remote than I thought it would be and I have had to refresh my knowledge in this area. It’s been quite hard since I started to get into the field, but I’m getting there,” he declares.

He has also found that another advantage with being in the human medicine field is that there are so much more resources. In his previous studies of ecology and corals, funding was more scarce and therefore the number of researchers involved was smaller, making the work more lonely.

Potential additional collaborations

So far Hadrien has only stayed in touch with his previous group at SLU to finalise some of his work there. When he left, it was decided that he should return to give a presentation about the post doc project but due to the pandemic this has not been realised yet.

“I really hope that this can be arranged in the not-too-distant future. If not for the pandemic, I believe that I would have had many more interactions with my old department,” he says.  

They also hope to expand their work on bringing the machine learning competence from SLU into medical research. There is an increasing interest in using AI in medical applications and they have discussed collaboration with another at research group at the Medfarm Domain. At present, however, there are limitations on how much AI can be used.

“This type of research requires extremely large data sets and this doesn’t work in clinical environments where the interest usually focuses on studying single patients or small groups that are thoroughly characterised. So, the stage that we are working on now, where we are testing the methodology based on a lot of data, is not easily applied in medicine. But we hope that our tests can result in methods that can eventually have medical applications,” Åsa concludes.