Lecture 7 - Polygenic additive models and Polygenic Risk Scores
Find lecture notes here
Summary of the notes by Gemini reviewed by Haky
Summary of Lecture Notes
The lecture primarily focuses on polygenic risk scores (PRS) and the statistical methods and challenges associated with their development and application. Key points include:
Polygenic Additive Model: The foundation of PRS is the polygenic additive model, which posits that complex traits are influenced by the sum of many genetic variants, each with a small effect in general.
Parameter Estimation Challenges: Estimating the effects of all SNPs simultaneously is challenging due to the large number of parameters and limited sample sizes.
Solutions to Parameter Estimation: The lecture discusses several strategies to address this, including modeling SNP effects as random effects, using GWAS effect sizes, and penalized likelihood methods (regularization) like Ridge, LASSO, and Elastic Net.
Alternative Prediction Approaches: Beyond simple PRS, the lecture touches on more complex methods like BSLMM, which model genetic effects as a mixture of normal distributions.
Improving PRS Accuracy: The importance of using Linkage Disequilibrium (LD) information to refine PRS calculations is emphasized, with methods like PRSice, LD-Pred, and others mentioned.
Deep Learning in Genomic Prediction: The potential of deep learning to improve genomic prediction is explored, but the lecture notes suggest that, so far, deep learning methods have not consistently outperformed linear models for PRS.
Clinical Utility and Limitations of PRS: The lecture discusses the potential clinical utility of PRS in predicting disease risk, using examples like height, coronary artery disease, and type 2 diabetes. However, it also highlights limitations, particularly the reduced accuracy of PRS across diverse ancestries.
Ethical Considerations: The lecture concludes by addressing the bioethical issues surrounding the use of PRS in embryo screening, an active area of research and debate.