QTG-Finder2: A Generalized Machine-Learning Algorithm for Prioritizing QTL Causal Genes in Plants.

Area of Science:

• Plant genetics
• Bioinformatics
• Machine learning

Background:

• Quantitative trait loci (QTL) mapping is crucial for identifying genes controlling complex traits in plants.
• Fine-mapping QTL to pinpoint causal genes is often time-consuming and labor-intensive.
• Previous tools like QTG-Finder were limited to species with extensive known causal gene data.

Purpose of the Study:

• To develop a machine-learning approach (QTG-Finder2) for identifying causal genes in plant QTLs, even in species with limited known genetic information.
• To enhance QTG-Finder's capabilities by incorporating orthologs and genetic variations.
• To accelerate the discovery of genes responsible for important agricultural traits.

Main Methods:

• Utilized orthologs of known causal genes as a training set for machine learning models.
• Extended the algorithm to include polymorphisms in conserved non-coding sequences and gene presence/absence variation.
• Trained and cross-validated QTG-Finder2 models for *Sorghum bicolor* and *Setaria viridis*.
• Validated the *S. bicolor* model with literature-curated causal genes and applied the *S. viridis* model to a plant height QTL.

Main Results:

• Models trained with orthologs achieved comparable performance to those trained with known causal genes (e.g., 83% recall for rice).
• QTG-Finder2 demonstrated high recall rates (70% for *S. bicolor* top 20% ranked genes).
• Identified 13 candidate genes for a plant height QTL in *S. viridis*.

Conclusions:

• QTG-Finder2 effectively prioritizes candidate causal genes in QTLs across diverse plant species, including those with limited genomic data.
• The inclusion of orthologs and genetic variations significantly broadens the applicability of the tool.
• QTG-Finder2 accelerates the identification of genes underlying important agricultural traits, facilitating crop improvement.