EAT-Rice

EAT-Rice predicts changes in expression of genes flanking T-DNA insertion sites in rice mutants to support functional characterization of genes altered by T-DNA activation-tagging with the CaMV 35S enhancer.

Key Features:

• Two-layer machine learning framework: The first layer comprises Support Vector Machine (SVM) models that analyze three DNA sequence types—UPS1K, DISTANCE, and MIDDLE—encoded using N-gram, Motif, nucleotide physicochemical properties (NPC), and CG-island (CGI) features, with logistic regression applied to estimate activation probability.
• Integration with NaiveBayesUpdateable: The second layer integrates outputs from the first-layer SVM/logistic regression models using the NaiveBayesUpdateable algorithm to improve predictive accuracy.
• Performance metrics: Reported accuracy is 88.33% on 5-fold cross-validation and 79.17% on independent testing, with the MIDDLE sequence type contributing most to activation identification.
• Comparative advantage: Outperforms the TRIM database by providing more accurate predictions of gene expression at greater distances from T-DNA insertion sites.

Scientific Applications:

• Functional genomics and gene identification: Predicts flanking gene expression changes to accelerate identification and analysis of genes affected by T-DNA insertions in rice.
• Experimental validation prioritization: Produces probability-based predictions to prioritize candidate activated genes and reduce the scope of biological validation experiments.
• Regulatory mechanism and network analysis: Supports investigation of cis-regulatory element effects and genetic networks altered by CaMV 35S activation-tagging in rice.

Methodology:

Retrieve DNA sequences around T-DNA insertion sites; encode sequences using N-gram, Motif, NPC, and CGI features; train SVM models on UPS1K, DISTANCE, and MIDDLE sequences; apply logistic regression for probability estimation; and integrate first-layer outputs with NaiveBayesUpdateable.