Improving genomic selection accuracy using a dual-path convolutional neural network framework: a terpenoid case study

Area of Science:

• Plant genetics and breeding
• Genomics
• Metabolomics

Background:

• Genomic selection (GS) is crucial for accelerating early breeding in forest trees.
• Understanding the genetic basis of terpenoid biosynthesis is key for improving valuable traits in species like Litsea cubeba.
• Traditional breeding methods can be slow, necessitating advanced genomic tools for efficient genetic improvement.

Purpose of the Study:

• To investigate the genetic architecture of terpenoid biosynthesis in Litsea cubeba.
• To develop and evaluate a novel deep learning-based genomic selection (GS) model (PKDP) for enhanced prediction accuracy.
• To identify key genes and pathways involved in terpenoid synthesis for molecular breeding.

Main Methods:

• Whole-genome resequencing of 945 Litsea cubeba germplasms.
• Gas Chromatography-Mass Spectrometry (GC-MS) for terpenoid quantification in 310 samples.
• Genome-wide association studies (GWAS) and development of the PKDP deep learning model integrating GWAS loci and genome-wide markers.

Main Results:

• GWAS identified 125 candidate genes for terpenoid biosynthesis, including terpene synthase (TPS) gene clusters and MVA/MEP pathway enzymes.
• The PKDP deep learning model demonstrated improved predictive ability for key terpenoids (citral, geranial, neral) by 2-10% compared to traditional rrBLUP.
• Key genetic factors influencing Litsea cubeba terpenoid synthesis were elucidated.

Conclusions:

• The study identified critical genetic factors for Litsea cubeba terpenoid synthesis, advancing molecular mechanism research.
• The PKDP model effectively integrates multi-scale genomic information, significantly enhancing GS prediction accuracy.
• This research provides an efficient tool for Litsea cubeba genetic improvement and offers innovative strategies for breeding complex traits in forest trees.