Optimizing genomic prediction with transfer learning under a ridge regression framework.

Area of Science:

• Agricultural Science
• Genetics
• Biotechnology

Background:

• Genomic selection (GS) is crucial for predicting plant and animal traits without direct phenotyping.
• Improving prediction accuracy in GS is vital due to experimental noise.
• Various strategies exist to enhance GS prediction accuracy.

Purpose of the Study:

• To explore the application of transfer learning in genomic selection.
• To evaluate transfer learning combined with ridge regression (RR) and analytic RR (ARR) for enhanced prediction.
• To compare transfer learning models against traditional RR and ARR.

Main Methods:

• Applied transfer learning (Transfer RR and Transfer ARR) from proxy to goal environments.
• Utilized 11 multi-environment datasets for wheat and rice.
• Evaluated model performance using Pearson's correlation (Cor) and normalized root mean square error (NRMSE).

Main Results:

• Transfer RR and Transfer ARR significantly improved predictive performance.
• The transfer learning approaches enhanced Cor by 22.962% and NRMSE by 5.757% compared to non-transfer models.
• Empirical evidence supports the effectiveness of transfer learning in GS.

Conclusions:

• Transfer learning, specifically Transfer RR and Transfer ARR, offers a powerful approach to boost genomic selection accuracy.
• These methods demonstrate significant potential for improving breeding programs.
• The findings highlight the value of leveraging data across environments for more accurate predictions.