WARNING: the interactive features of this website use CSS3, which your browser does not support. To use the full features of this website, please update your browser.
Neural networks have been seldomly leveraged in population genomics due to the computational burden and challenge of interpretability. Here, we propose GenNet, a novel open-source deep learning framework for predicting phenotype from genotype. In this framework, public prior biological knowledge is used to construct interpretable and memory-efficient neural network architectures. These architectures obtain good predictive performance for multiple traits and complex diseases, opening the door for neural networks in population genomics.
Genome-wide and brain-wide association studies
Genome-wide association studies (GWAS) have identified dozens of genetic variants related to brain structure. These studies have focused on a few aggregate measures since GWAS are computationally demanding. However, the brain is complex and can be described using millions of measures. We recently developed methods to perform such genome-wide and brain-wide association studies and applied this in several cohort studies. We would like to replicate our findings in the UK Biobank by jointly meta-analyzing the results. It fits well with the purpose of the UK Biobank in two ways:
1) It builds upon this major resource by providing a range of novel neuroimaging biomarkers, which will be made available to other researchers
2) It will hopefully identify novel genetic variants that are important for brain structure and diseases of the brain (e.g., Alzheimer's disease, schizophrenia)
We will analyze brain images to calculate million of measures that describe the structure of the brain. Next, we will perform genome-wide screens of millions of genetic variants to identify ones that affect brain structure. The full cohort of individuals with both brain imaging and genetic data available.