Differential Gene Expression Prediction by Ensemble Deep Networks on Histone Modification Data.

Area of Science:

• Genomics
• Epigenetics
• Computational Biology

Background:

• Histone modifications (HM) regulate gene expression, influencing cell heterogeneity.
• Predicting differential gene expression (DGE) from HM signals is vital for understanding gene regulation.
• Current methods using fixed-length bins and single models face limitations like information loss and model bias.

Purpose of the Study:

• To develop an advanced framework for predicting differential gene expression from histone modification signals.
• To overcome limitations of fixed-length binning and single model biases in DGE prediction.
• To improve the accuracy and robustness of DGE prediction in single-cell studies.

Main Methods:

• Proposed an Ensemble deep neural networks framework (EnDGE) for DGE prediction.
• Utilized diverse feature extractors with varying bin lengths for HM signal processing.
• Integrated a novel Residual Network-based model to enhance feature extraction diversity.
• Employed ensemble learning with varied signal segmentation strategies to mitigate individual model biases.

Main Results:

• EnDGE demonstrated significantly superior performance compared to state-of-the-art baselines across all cell type pairs.
• The ensemble approach effectively preserved the integrity and consistency of genetic information.
• The novel Residual Network improved prediction accuracy and extractor diversity.

Conclusions:

• EnDGE offers a robust and accurate method for predicting differential gene expression from histone modification data.
• The framework's ensemble strategy and diverse feature extraction enhance biological information capture.
• This approach advances our understanding of epigenetic regulation in cell functional heterogeneity.