Meta-i6mA

Meta-i6mA predicts DNA N6-methyladenine (6mA) sites across plant genomes using an integrative machine-learning meta-predictor.

Key Features:

• 6mA site prediction: Predicts DNA N6-methyladenine (6mA) sites across diverse plant genomes.
• Feature encoding evaluation: Explored 10 distinct feature encoding schemes and selected five schemes based on physicochemical and position-specific information.
• Machine learning algorithms: Utilizes six classifiers: random forest, support vector machine, extremely randomized tree, logistic regression, naïve Bayes, and AdaBoost.
• Training data and models: Trained on the Rosaceae genome to generate 30 baseline models.
• Meta-predictor integration: Integrates the 30 baseline models into a meta-predictor to combine strengths of individual classifiers.
• Performance: Achieved Matthews correlation coefficient (MCC) values of 0.918 for Rosaceae, 0.827 for rice, and 0.635 for Arabidopsis thaliana in independent testing.

Scientific Applications:

• Cross-species 6mA prediction: Enables prediction of 6mA sites across multiple plant species, demonstrated on Rosaceae, rice, and Arabidopsis thaliana.
• Epigenetic research: Supports study of epigenetic regulation by identifying candidate 6mA sites for functional characterization.

Methodology:

Explored 10 feature encoding schemes, selected five based on physicochemical and position-specific information, trained six classifiers (random forest, support vector machine, extremely randomized tree, logistic regression, naïve Bayes, AdaBoost) on the Rosaceae genome to produce 30 baseline models, and integrated them via a meta-predictor; performance was assessed by independent testing reporting MCC values for Rosaceae, rice, and Arabidopsis thaliana.