A Pipeline for Classifying Deleterious Coding Mutations in Agricultural Plants.

Area of Science:

• Plant genetics
• Bioinformatics
• Evolutionary biology

Background:

• Deleterious genetic variations impact plant fitness and crop productivity, but their prediction is challenging.
• Current methods rely on sequence conservation, lacking function-based classifiers due to limited annotated mutation data in plants.

Purpose of the Study:

• To develop and evaluate a machine learning classifier for accurate prediction of deleterious mutations in plants.
• To assess the classifier's performance across different plant species and compare it with existing tools.

Main Methods:

• Extracted 18 features, including 9 novel ones, to train machine learning models (SVM, Random Forest) on Arabidopsis thaliana mutation data.
• Evaluated Random Forest classifier performance on Oryza sativa, Pisum sativum, and Cicer arietinum datasets.
• Compared classifier accuracy against the PolyPhen-2 tool and utilized population frequency data for validation.

Main Results:

• The Random Forest classifier demonstrated superior accuracy in predicting deleterious mutations in Arabidopsis thaliana compared to PolyPhen-2.
• The classifier achieved high prediction accuracies (87% and 93%) for Oryza sativa and Pisum sativum, respectively.
• Transfer learning did not enhance classifier performance; however, application to Cicer arietinum showed promising results validated by population data.

Conclusions:

• A robust machine learning classifier for identifying deleterious plant mutations has been developed.
• This tool can enhance the annotation of functional mutations in quantitative trait loci (QTL) and genome-wide association study (GWAS) regions.
• The classifier offers potential for evolutionary analyses and optimizing plant breeding for improved cultivars.