Deep Learning Algorithms for Complex Traits Genomic Prediction

The underlying perception of genomic selection (GS) is to use genome-wide from DNA sequence (“SNP markers”) along with phenotypes an observed population make prediction for the phenotypic outcomes untested individuals in crop and livestock breeding programs. GS was firstly described by Meuwissen et al.(2001) dairy cattle identify genetically superior animals at early age. aim capture specific genes across whole genome that are associated desired traits. major challenge using programs predict effect many SNP markers information a few (aka small n big p problem, or >> n). Many approaches including naïve scaled elastic net, ridge regression BLUP Bayesian (BayesA, BayesB, BayesCπ, BayesDπ) LASSO, Support Vector Regression have been conducted address (aka, n) problem. These methods all perform well (p>>n) linear approximation set functional relationship between genotypes phenotypes. However, these may not fully non-linear effects which possible be crucial complex To deal this limitation, neural networks (NN) were recommended cover non-linearity GS. Artificial NNs (ANNs) first presented Okut al. (2011) who establish connected regularized multi-layer ANN (MLANN) comprising one hidden layer body mass index (BMI) mice dense molecular markers. Since then, rather ANNs applied deep learning (DL) networks. different DL algorithms their own advantages problems trait Four classes such as artificial (DL-MLANN), recurrent (RNN), convolutional (CNN) long-short term memory (LSTM) some variation network architectures will summarized here.