Abstract:
|
Both genetic factors and lifestyle confer risk for breast cancer. However due to lack of accurate genetic risk estimate caused by difficulty in estimating the effect size of SNPs and in identifying the most functionally relevant SNPs, interactions between genetic factors and lifestyle for breast cancer were not fully understood. In this research, we propose an integrated model aiming to address these challenges by employing AnnoPred, a Bayesian framework that leverages functional annotations in quantifying and enhancing polygenic risk score (PRS). We applied this model to 227,223 (17,027 breast cancer patients) unrelated British participants. K-fold AUC cross validation on the enhanced PRS revealed improved genetic prediction performance of AnnoPred compared with state-of-the-art PRS methods. We then analyzed the compounding interactions between genetic and non-genetic lifestyle factors on breast cancer incident rate and observed a strong positive correlation between PRS and breast cancer incident rate. We also found that adherence to a healthy lifestyle reduces breast cancer risk. Lastly, we combined PRS and lifestyle factors to train a MLP neural network, achieving a 93% accuracy in predicting breast cancer.
|