Breast cancer is the most common cancer in women in the US and in the world. Stratification of women according to the risk of developing breast cancer could improve risk reduction and screening strategies by targeting those most likely to benefit. Recently, polygenic risk scores (PRSs) have been developed to predict breast cancer in Caucasian and Asian populations. However, there are no PRSs that perform well for African Americans. To predict breast cancer risk in AAs, existing PRS methods often train PRSs using summary statistics from European-based studies and apply the PRSs to AAs. Due to distinct allele frequencies and linkage disequilibrium (LD) structures across populations, the prediction accuracy is quite low. There is a need to construct risk scores using AA training data to improve the prediction accuracy. African Americans are an admixed population formed by recent admixture of mainly two ancestral populations: West Africans and European Americans. The admixture between dif- ferent continental populations creates mosaic chromosomes that contain long segments of distinct ancestry and therefore induces admixture LD (ALD) among markers in large chromosomal regions (with several Mbs). The local ancestries at different markers in an ALD region may provide useful information to tag the effect of some causal variants in the same ALD region. This information is supplementary to that used in traditional PRSs. To our knowledge, existing PRSs for breast cancers in AAs often do not model local ancestry and ALD. In addition, recently, function annotations based on genomic and epigenomic data have been incorporated into PRSs to improve prediction accuracy. The objective of this study is to construct effective genetic risk scores for prediction of breast cancer in AA women. Specifically, we will propose a novel application of two Bayesian PRS methods, LDpred and AnnoPred, to AA data and construct joint risk prediction scores that integrate traditional PRSs, local ancestry, and genomic and epigenomic functional annotations based on effectively modeling ALD. We will apply the joint risk scores to analyses of real African American breast cancer data and evaluate the risk scores by simulation studies. We will train the joint risk scores using AA samples to improve prediction accuracy. The proposed joint risk prediction models have a good potential to translate knowledge from GWAS discovery to the practice of breast cancer screening and advance the science of precision prevention. The proposed joint risk scores are innovative because they can effectively model local ancestry and admixture LD while incorporating information from traditional PRSs and functional annotations. We expect that the proposed joint risk scores will have higher accuracy than traditional approaches in breast cancer predication in AA women. The proposed method can also be used or extended to predict other disease traits in admixed populations.
The proposed risk prediction methods have a good potential to translate knowledge from genome-wide association studies to the practice of breast cancer screening and advance the science of precision prevention in African American women.
