eRFSVM: a hybrid classifier to predict enhancers-integrating random forests with support vector machines

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Enhancers are critical tissue-specific elements regulating gene expression.
• Predicting enhancers is a significant bioinformatics challenge.
• Current methods have limitations in predicting EP300 enhancers and generalization.

Purpose of the Study:

• To develop an improved computational method for enhancer prediction.
• To enhance the accuracy and generalization of enhancer identification.

Main Methods:

• Developed a hybrid classifier, eRFSVM, combining Random Forests and Support Vector Machines.
• Integrated two components: eRFSVM-ENCODE and eRFSVM-FANTOM5, utilizing diverse features and labels.
• Trained base classifiers on single-tissue/cell data and main classifiers on their predictions.

Main Results:

• eRFSVM-ENCODE achieved 83.69% precision on K562 data, outperforming existing classifiers.
• eRFSVM-FANTOM5 demonstrated significant improvements: 86.17% precision, 36.06% recall, 50.84% F-score, and 93.38% accuracy.
• The hybrid approach showed superior performance in predicting both EP300 and FANTOM5 RNA-based enhancers.

Conclusions:

• eRFSVM is a highly effective classifier for predicting EP300-based enhancers.
• eRFSVM demonstrates superior performance in predicting FANTOM5 RNA-based enhancers.
• The hybrid eRFSVM classifier offers enhanced accuracy and generalization for enhancer identification.