EBMGP: a deep learning model for genomic prediction based on Elastic Net feature selection and bidirectional encoder representations from transformer's embedding and multi-head attention pooling DOI Creative Commons

Lu Ji,

Wei Hou, Heng Zhou

и другие.

Theoretical and Applied Genetics, Год журнала: 2025, Номер 138(5)

Опубликована: Апрель 19, 2025

Enhancing early selection through genomic estimated breeding values is pivotal for reducing generation intervals and accelerating programs. Recently, deep learning (DL) approaches have gained prominence in prediction (GP). Here, we introduce a novel DL framework GP based on Elastic Net feature bidirectional encoder representations from transformer's embedding multi-head attention pooling (EBMGP). EBMGP applies the of features, thereby diminishing computational burden bolstering predictive accuracy. In EBMGP, SNPs are treated as "words," groups adjacent with similar LD levels considered "sentences." By applying transformers embeddings, this method models manner analogous to human language, capturing complex genetic interactions at both "word" "sentence" scales. This flexible representation seamlessly integrates into any network demonstrates marked improvement performance SoyDNGP compared widely used one-hot representation. We propose pooling, which can adaptively assign weights features while multiple subspaces multi-heads high level semantic understanding. comprehensive comparative analysis across four diverse plant animal datasets, outperformed competing 13 out 16 tasks, achieving accuracy gains ranging 0.74 9.55% over second-best model. These results underscore EBMGP's robustness highlight its potential applications life sciences.

Язык: Английский

EBMGP: a deep learning model for genomic prediction based on Elastic Net feature selection and bidirectional encoder representations from transformer's embedding and multi-head attention pooling DOI Creative Commons

Lu Ji,

Wei Hou, Heng Zhou

и другие.

Theoretical and Applied Genetics, Год журнала: 2025, Номер 138(5)

Опубликована: Апрель 19, 2025

Enhancing early selection through genomic estimated breeding values is pivotal for reducing generation intervals and accelerating programs. Recently, deep learning (DL) approaches have gained prominence in prediction (GP). Here, we introduce a novel DL framework GP based on Elastic Net feature bidirectional encoder representations from transformer's embedding multi-head attention pooling (EBMGP). EBMGP applies the of features, thereby diminishing computational burden bolstering predictive accuracy. In EBMGP, SNPs are treated as "words," groups adjacent with similar LD levels considered "sentences." By applying transformers embeddings, this method models manner analogous to human language, capturing complex genetic interactions at both "word" "sentence" scales. This flexible representation seamlessly integrates into any network demonstrates marked improvement performance SoyDNGP compared widely used one-hot representation. We propose pooling, which can adaptively assign weights features while multiple subspaces multi-heads high level semantic understanding. comprehensive comparative analysis across four diverse plant animal datasets, outperformed competing 13 out 16 tasks, achieving accuracy gains ranging 0.74 9.55% over second-best model. These results underscore EBMGP's robustness highlight its potential applications life sciences.

Язык: Английский

Процитировано

0