Meeting 59, 2024-05-17

NBIS arranges an open AI and IO Seminar Series aimed at knowledge-sharing about Artificial Intelligence and Integrative Omics (AI & IO) analysis and applications in the Life Science community. The seminar series is open to everyone. The seminar is run over Zoom on the third Friday of the month during academic terms, typically between 10 and 11 am, with approx. 45 min long presentation and 15 min discussion. 

Abstract: Precision medicine is an important part of our future healthcare, since it allows you to predict what will happen in a specific patient. Much attention is given to machine learning (ML), which has important showcases regarding e.g. multi-omics and image analysis. Such ML-based models are trained on large-scale datasets, with thousands of patients, which have been characterized at one or a few timepoints. However, such databases almost always lack the information about what the person has been doing in-between those timepoints. Without adding this lacking information into the models, one cannot say how a particular patient is going to evolve, but only how an average person with those characteristics is going to evolve. In this talk, I will outline how we use a hybrid modelling approach – digital twins – to add this vital information. The first step is to integrate the required data, available from different sources, by storing a copy of all data in a personal data vault. This idea overcomes most legal and ethical problems associated with the traditional creations of large-scale databases. The second step is to do hybrid data analysis combining ML and mechanistic modelling. More specifically, hypothesis testing and mechanistic models are used to predict how risk factors evolve in response to what a person is doing, and those simulations are then used as inputs to statistical ML-models. The resulting digital twin models range from intracellular processes (multi-omics, signalling, metabolism, gene networks), to organ-organ cross-talk, and whole-body disease development. Finally, I will give examples of how we these digital twins, which can visualize simulation results by an avatar that looks like the patient, in real healthcare, currently tested in the EU-consortium STRATIF-AI.

Short bio for the speaker: Gunnar Cedersund heads a research group of ~25 people at Linköping university, consisting of mathematical modelers, experimentalists, medical doctors, software engineers, and machine learning experts. Together they have spent the last 20+ years to develop a unique technology: multi-organ and multi-timescale digital twins. His digital twins are a computer copy of a person, which looks like that person on the outside (face, body type, etc), and on the inside (organs, cells, etc). The digital twins can simulate what happens if the person e.g. changes their diet, starts to exercise, or takes certain medications. These digital twins are currently being tested in 300+ patients, in the big EU consortium STRATIF-AI, which is coordinated by Cedersund, and which uses digital twins and AI to aid with prevention, acute treatment and rehabilitation of stroke. Apart from this, Cedersund has funding from e.g. VR-M, VR-NT, KAW, VINNOVA, SSF, AstraZeneca, and many other sources.

Slides: