| Title: | Leveraging functional genomics resources to improve the power of phenotype prediction -- a new machine learning approach |
| Lead Institution: | Beijing Genomics Institute At Shenzhen |
| Principal investigator: | Dr Jiang Xiaosen |
Application 65814
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.
About
AIMS 1. We have developed a new set of deep learning framework, abandoning the traditional prediction based on single SNP locus information, we have carried out SNP information with the gene location, chromosome segment information, functional pathway, etc. Combine them as a feature for training. At present, the test results are good for our low-depth data. We hope to apply for UKB data and test on high-depth data. 2. If the test results are good, we hope to expand the method and combine the results of the method with other omics data to comprehensively predict the phenotype or disease risk. 3. We hope that this method will be made into a set of general calculation framework for everyone to use 4. If the model is successfully trained, we can classify the populations based on the scores, compare the differences between the populations, and find key pathways or site information. Scientific basis: 1. The existing polygenic risk assessments have achieved some very good results, which reflects the feasibility of this direction. 2. There are reports in the literature about methods for multi-gene risk assessment based on deep learning, but the accuracy and stability are not as good as traditional methods. 3. SNPs are not completely independent. They are connected to chromosome positions and functional pathways. This connection may be a non-linear relationship. This relationship is difficult to capture in traditional prediction methods. What we have constructed The model can reflect these relationships in the features, which may improve the performance of prediction. 3. At present, there are many attempts to conduct association analysis based on multi-omics data, and good results have been achieved. We can also introduce it into our research and make the model more biologically explanatory through multi-omics data. project duration! ~ 36 months public health impact: 1. Hope that our project can discover potentially key genetic information in some diseases or phenotypes and help clinical treatment. 2. I hope that the results of our project can be put into the clinic, help doctors predict diseases and help police predict phenotypes...4 Publications
| Pub ID | Title | Author(s) | Year | Journal |
|---|---|---|---|---|
| 10174 | Cystatin C-based estimated glomerular filtration rate and risk of stroke in the general population: a prospective cohort study | Jinlan Liao (+17) | 2023 | Clinical Kidney Journal |
| 10982 | Defining reference values for body composition indices by magnetic resonance imaging in UK Biobank | Liang Dai (+9) | 2023 | Journal of Cachexia Sarcopenia and Muscle |
| 9074 | Kidney Function Measures and Mortality: A Mendelian Randomization Study | Ying Shan (+13) | 2023 | American Journal of Kidney Diseases |
| 11097 | Sex-specific equations to estimate body composition: Derivation and validation of diagnostic prediction models using UK Biobank | Yueqi Lu (+13) | 2023 | Clinical Nutrition |
Enabling scientific discoveries that improve human health