Portal de Periódicos da CAPES

Deep integrative models for large-scale human genomics

2023; Oxford University Press; Volume: 51; Issue: 12 Linguagem: Inglês

10.1093/nar/gkad373

1362-4962

Arnór I. Sigurdsson, Ioannis Louloudis, Karina Banasik, David Westergaard, Ole Winther, Ole Lund, Sisse Rye Ostrowski, Christian Erikstrup, Ole Birger Pedersen, Mette Nyegaard, Karina Banasik, Jakob Thaning Bay, Jens Kjærgaard Boldsen, Thorsten Brodersen, Søren Brunak, Kristoffer Sølvsten Burgdorf, Mona Ameri Chalmer, Maria Didriksen, Khoa Manh Dinh, Joseph Dowsett, Christian Erikstrup, Bjarke Feenstra, Frank Geller, Daníel F. Guðbjartsson, Thomas Hansen, Lotte Hindhede, Henrik Hjalgrim, Rikke Louise Jacobsen, Gregor B. E. Jemec, Kathrine Agergård Kaspersen, Bertram Kjerulff, Lisette J. A. Kogelman, Margit Anita Hørup Larsen, Ioannis Louloudis, Agnete Troen Lundgaard, Susan Mikkelsen, Christina Mikkelsen, Kaspar René Nielsen, Janna Nissen, Mette Nyegaard, Sisse Rye Ostrowski, Ole Birger Pedersen, Alexander Pil Henriksen, Palle Duun Rohde, Klaus Rostgaard, Michael Schwinn, Kāri Stefánsson, Hreinn Stefónsson, Erik Sørensen, Unnur Þorsteinsdóttir, Lise Wegner Thørner, Mie Topholm Bruun, Henrik Ullum, Thomas Werge, David Westergaard, Søren Brunak, Bjarni J. Vilhjálmsson, Simon Rasmussen,

Biomedical Text Mining and Ontologies

Polygenic risk scores (PRSs) are expected to play a critical role in precision medicine. Currently, PRS predictors are generally based on linear models using summary statistics, and more recently individual-level data. However, these predictors mainly capture additive relationships and are limited in data modalities they can use. We developed a deep learning framework (EIR) for PRS prediction which includes a model, genome-local-net (GLN), specifically designed for large-scale genomics data. The framework supports multi-task learning, automatic integration of other clinical and biochemical data, and model explainability. When applied to individual-level data from the UK Biobank, the GLN model demonstrated a competitive performance compared to established neural network architectures, particularly for certain traits, showcasing its potential in modeling complex genetic relationships. Furthermore, the GLN model outperformed linear PRS methods for Type 1 Diabetes, likely due to modeling non-additive genetic effects and epistasis. This was supported by our identification of widespread non-additive genetic effects and epistasis in the context of T1D. Finally, we constructed PRS models that integrated genotype, blood, urine, and anthropometric data and found that this improved performance for 93% of the 290 diseases and disorders considered. EIR is available at https://github.com/arnor-sigurdsson/EIR.