iEnhancer-XG: interpretable sequence-based enhancers and their strength predictor.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Enhancers are crucial non-coding DNA elements regulating gene expression.
• Current machine learning tools for enhancer identification have limitations in accuracy and efficiency.

Purpose of the Study:

• To develop an accurate and efficient predictor for identifying enhancers and their strengths.
• To improve the interpretability and credibility of machine learning models in genomics.

Main Methods:

• Proposed a two-layer predictor, iEnhancer-XG, utilizing XGBoost as the base classifier.
• Employed five feature extraction methods: k-Spectrum Profile, Mismatch k-tuple, Subsequence Profile, PSSM, and PseDNC.
• Integrated an ensemble learning approach for feature fusion and SHapley Additive explanations for model interpretability.

Main Results:

• Achieved prediction accuracies of 0.811 for enhancer identification (first layer) and 0.657 for strength prediction (second layer).
• Demonstrated superior performance compared to existing technologies through rigorous 10-fold cross-validation.

Conclusions:

• iEnhancer-XG offers a significant advancement in enhancer prediction accuracy and efficiency.
• The ensemble learning approach and interpretability methods enhance the reliability of computational predictions in genomics.