Machine learning and bioinformatics
The ISB group employs a versatile and hybrid approach to modeling, and we also have a subgroup devoted to machine learning and bioinformatics large-scale models. This sub-group works on a variety of different approaches. Rasmus Magnusson and Hendrick de Weerd have pioneered a new approach to analyze large-scale gene expression data, using something called variational autoencoders (VAE) (de Weerd et al, 2024). This approach has allowed us to extract the information from millions of experiments that have measured gene expression (mRNA) in a variety of different contexts. Even though these different experiments have been done in different conditions, we can use VAE to identify a smaller latent space - the space in which genes move, while taking all covariations into account. This approach can then be used to analyze e.g. specific small-scale data from a specific lab, such as gene expression data from cancer patients studied in the lab of Linda Bojmar, who study the existence of a pre-metastatic niche. Another application of such methodologies is analysis of single-cell gene expression leukemia data, studied in the lab of Mikael Sigvardsson.
We also have growing research activities in the field of metabolomics modeling. On the cellular level, this involves analysis of mass spectrometry data, which measures the distribution of labelled 13C atoms in intracellular samples. This has allowed us to characterize intracellular metabolic fluxes in liver tissue, as recently published in Nature Metabolism (Grankvist et al, 2024). On the patient level, we also analyze mass spec data from blood samples. This is done to identify biomarkers, which can be used either to characterize e.g. liver patients, or to identify when and by which cause a diseased person has died. The latter is part of a big forensic project financed by the Swedish Reserch Council.
Finally, we also do a lot of hybrid modeling, which combines machine learning and mechanistic modeling in different ways. We have e.g. introduced a way to automatically scale mechanistic models to the omics level, and applied this both to the analysis of gene expression data and to the identification of gene networks (Magnusson, 2017), and to the analysis of phosphoproteomics data and the identification of protein signaling networks (Lövfors, 2023). We have also developed a method to combine mechanistic simulations with machine learning models, and used this to simulate our digital twin models: the resulting hybrid models can simulate not only physiological variables, such as glucose, blood pressure, and weight, but also the resulting statistical risk of a stroke (Herrgårdh, 2021;Herrgårdh, 2022).
Henrik A de Weerd, Dimitri Guala, Mika Gustafsson, Jane Synnergren, Jesper Tegnér, Zelmina Lubovac-Pilav, Rasmus Magnusson (2024). Latent space arithmetic on data embeddings from healthy multi-tissue human RNA-seq decodes disease modules, Patterns (N Y), 5(11):101093. Take me to the article
Nina Grankvist, Cecilia Jönsson, Karin Hedin, Nicolas Sundqvist, Per Sandström, Bergthor Björnsson, Arjana Begzati, Evgeniya Mickols, Per Artursson, Mohit Jain, Gunnar Cedersund, Roland Nilsson (2024). Global 13C tracing and metabolic flux analysis of intact human liver tissue ex vivo, Nat Metab., 6(10):1963-1975. Take me to the article
Rasmus Magnusson, Guido Pio Mariotti, Mattias Köpsén, William Lövfors, Danuta R Gawel, Rebecka Jörnsten, Jörg Linde, Torbjörn E M Nordling, Elin Nyman, Sylvie Schulze, Colm E Nestor, Huan Zhang, Gunnar Cedersund, Mikael Benson, Andreas Tjärnberg, Mika Gustafsson (2017). LASSIM-A network inference toolbox for genome-wide mechanistic modeling, PLoS Comput Biol., 13(6):e1005608. Take me to the article
William Lövfors, Rasmus Magnusson, Cecilia Jönsson, Mika Gustafsson, Charlotta S Olofsson, Gunnar Cedersund, Elin Nyman (2023). A comprehensive mechanistic model of adipocyte signaling with layers of confidence, NPJ Syst Biol Appl., 9(1):24. Take me to the article
Tilda Herrgårdh, Vince I Madai, John D Kelleher, Rasmus Magnusson, Mika Gustafsson, Lili Milani, Peter Gennemark, Gunnar Cedersund (2021). Hybrid modelling for stroke care: Review and suggestions of new approaches for risk assessment and simulation of scenarios, Neuroimage Clin., 31102694. Take me to the article
Tilda Herrgårdh, Elizabeth Hunter, Kajsa Tunedal, Håkan Örman, Julia Amann, Francisco Abad Navarro, Catalina Martinez-Costa, John D. Kelleher, Gunnar Cedersund (2022). Digital twins and hybrid modelling for simulation of physiological variables and stroke risk, bioRxiv, preprint. Take me to the article
Key personnel
Rasmus Magnusson
Dirk de Weerd
Ralph Monte
Elin Nyman
Gunnar Cedersund
Tilda Herrgårdh