Genomic language model-based genomic prediction in plant breeding

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Genomic prediction using molecular markers has improved genomic selection.
• Prediction accuracy often plateaus despite increased marker density and refined methods.
• This saturation limits the utility of current genomic information.

Purpose of the Study:

• To explore genomic language models (GLMs) as a novel framework for genomic prediction.
• To investigate the potential of GLMs in capturing biologically meaningful DNA sequence grammar.
• To address the limitations of traditional marker-based approaches in genomic prediction.

Main Methods:

• Utilizing genomic language models (GLMs) to process sequence-based genomic information.
• Comparing GLM performance against traditional marker-based genomic prediction methods.
• Evaluating the biological expressivity of genomic representations.

Main Results:

• Genomic language models (GLMs) provide a new framework for genomic prediction.
• GLMs can incorporate richer sequence-based information, capturing DNA sequence grammar.
• This approach has the potential to overcome the plateau in prediction accuracy.

Conclusions:

• The future of genomic prediction relies on the biological expressivity of genomic representations, not just algorithmic refinement.
• Genomic language models (GLMs) offer a principled method to expand this representational frontier.
• GLMs represent a promising advancement for improving genomic selection accuracy.