wenlong2023/deepannotation repository overview

Leveraging genomic information to predict the potential breeding value of individuals, genomic selection has drastically expedited the enhancement of diverse economic traits. To further catalyze this progress, the influx of extensive multi-omics data in agricultural species has provided unprecedented opportunities, enriching the process with prior biological information. However, effectively harnessing these rich data resources to accurately predict phenotypes in genomic selection is still in its nascent stages. In this study, we present a novel interpretable genomic selection model, named DeepAnnotation, designed for phenotype prediction by leveraging comprehensive multi-omics functional annotations through deep learning technology. To capture the complex information flow from genotype to phenotype, DeepAnnotation aligns multi-omics biological annotations with sequential network layers in deep learning architecture, resembling the natural regulatory information transmission from genotype to intermediate molecular phenotypes—such as cis-regulatory elements, genes, and gene modules—and ultimately to phenotypes of economic traits. Comparing against three classical models—rrBLUP, LightGBM, and KAML—this design showcased superior performance in identifying top-performing individuals and enhancing computational efficiency for pork production traits. Furthermore, the interpretability embedded in our framework allows us to pinpoint potential causal SNPs and propose their mediated molecular mechanisms underlying these traits. DeepAnnotation is an open-source approach that designed to predict phenotypes utilizing comprehensive multi-omics functional annotations.