Artikel
Deep learning for genetic association testing with image phenotypes
Suche in Medline nach
Autoren
-
Matthias Kirchler
- Hasso Plattner Institut für Digital Engineering, Potsdam, GermanyTU Kaiserslautern, Kaiserslautern, Germany
-
Shahryar Khorasani
- Hasso Plattner Institut für Digital Engineering, Potsdam, Germany
-
Stefan Konigorski
- Hasso Plattner Institut für Digital Engineering, Potsdam, Germany
-
Marius Kloft
- TU Kaiserslautern, Kaiserslautern, Germany
-
Christoph Lippert
- Hasso Plattner Institut für Digital Engineering, Potsdam, Germany
| Veröffentlicht: | 26. Februar 2021 |
|---|
Gliederung
Text
In biomedical research we often are interested in testing for statistical significance, for example, to test for significant differences between features of disease cases and control individuals, or to test for association between a genetic variant and a disease phenotype. While this methodology is well-studied for tabular data, it is less obvious how to perform statistical testing with images. In this talk, I will give an overview over our work on using convolutional neural networks for this task. While convolutional neural networks excel in supervised prediction tasks from images, where we are interested in predicting a label from the pixel values, they have been less used for statistical hypothesis testing. I will present a non-parametric two-sample test based on deep learning embeddings [Kirchler et al., 2020] and show some example applications of for genome-wide association studies of image phenotypes extracted from retinal fundus images.
The authors declare that they have no competing interests.
The authors declare that an ethics committee vote is not required.
