SDM6A

SDM6A: Sequence-Based Machine-Learning Predictor of DNA N6-Methyladenine Sites in Rice

SDM6A predicts DNA N6-adenine methylation (6mA) sites in the rice genome using a two-layer ensemble machine-learning framework based on multiple sequence-derived feature encodings.

Key Features:

• Two-Layer Ensemble Architecture: Integrates predictions from multiple classifiers to generate final 6mA site predictions.
• Multiple Encoding Methods: Extracts features from five distinct sequence-based encoding schemes.
• Machine-Learning Models: Employs support vector machines and extremely randomized trees to construct individual predictive models.
• Integrated Classifier Fusion: Combines outputs of five single models in the first layer and integrates two classifiers in the second layer for final classification.
• Performance Metrics: Achieves 88.2% average accuracy and a Matthews correlation coefficient (MCC) of 0.764, improving accuracy by 4.7%–11.0% and MCC by 2.3%–5.5% over existing methods.

Scientific Applications:

• Rice Epigenetics: Identifies putative 6mA sites to support studies of DNA N6-methyladenine-mediated gene regulation and expression in rice.
• Crop Research: Facilitates epigenetic marker discovery relevant to rice breeding and functional genomics.

Methodology:

SDM6A extracts sequence-derived features using five encoding methods and trains individual classifiers based on support vector machines and extremely randomized trees. The first ensemble layer integrates predictions from five single models by assigning class labels according to classifier outputs. The second layer combines two classifiers to produce the final 6mA site prediction. Performance is evaluated using cross-validation and independent test datasets.