Data-harmonization and Artificial Intelligence were the main topics of the 10th Youth-GEMs Science Café. First guest in this Science Café was Lauren Fromont, a neuroscientist en researcher at the European Genomic-phenomic Archive (EGA) in Barcelona. In her presentation she focused on the challenge of harmonizing different types of data. In our Youth-GEMs research we are working with different types of data that are supposed to have an impact on mental health. We take into account:
genetic factors like mutations in the genes;
- clinical factors like e.g. anxiety or executive functions;
- environmental factors like experiences, family situations and life events;
- and digital factors like mobility, physical activity or phone usage.
The question is how to analyse these over 100 variables that come from different sources and different formats. Therefore a task force is working on harmonizing these data, which takes a lot of time and patience since it is mostly done by hand. A data model, which is a set of rules to determine the format of and the relationships between the variables, is now defined. The variables are harmonised according to this model.
Next guest in this Science Café was Karim Lekader, Director of the Artificial Intelligence in Medicine Lab at the University of Barcelona. He asked the Young Experts to define AI. Together the following definition was formed: ‘AI can be defined as machines trying to perform tasks by learning through examples’. The machines learn by finding patterns in (loads of) examples and use mathematical functions to find those patterns. When there are a few parameters, a regression analysis can be used, but mostly a lot of parameters should be used to perform the task properly and deep learning is needed. The more examples and parameters provided, the more powerful AI is. Karim also asked the Young Experts whether AI is a hype or a hope. We concluded that is a good thing on the condition that is used in a controlled and ethically sound way.
AI can be used in mental health to make diagnoses, predict outcomes of treatments or predict risks of suicide. However the challenges are amongst others:
- What if a person is not (much) online
- How and when do we communicate predicted risks
- These data are subjective (and not objectively measurable)