IFAM: Improving genomic prediction accuracy of complex traits by integrating massive types of functional annotation information

Genomic prediction which makes use of genome-wide genetic markers to predict complex traits had made great achievements during the past decade. With the development of omics techniques, the number of functional genomic annotations increased significantly, and leveraging this information in statistical models can potentially improve prediction performance. However, to effectively utilize the vast variety of functional annotations still faces big challenges. Herein, we developed an adaptive model named ‘IFAM’, which extends the linear mixed model with multiple random effects to accommodate massive types of functional annotations to improve the genomic prediction accuracy for complex traits. The IFAM yielded notable improvements on prediction accuracy across 20 traits from diverse datasets compared with the baseline GBLUP model. Briefly, IFAM achieved an average improvement of 9.43%, 6.25%, and 4.61% at the WTCCC1, UK Biobank, and pig datasets, respectively. Our findings highlight the effectiveness of integrating functional annotations to improve accuracy of genomic predictions.