DeepHEM: A novel deep domain-adversarial learning framework for identifying human essential miRNAs.

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Identifying essential human microRNAs (miRNAs) is crucial but challenging due to limited experimental data.
• Existing cross-species domain adaptation methods struggle with complex features and high-dimensional data.

Purpose of the Study:

• To develop an effective deep learning framework, DeepHEM, for predicting human miRNA essentiality.
• To improve cross-species knowledge transfer for miRNA function prediction.

Main Methods:

• Proposed DeepHEM, a deep domain-adversarial learning framework.
• Utilized a multi-modal feature extractor for sequence data, inherent properties, and miRNA-target gene interactions.
• Employed multi-loss optimization for feature alignment and domain-invariant learning.

Main Results:

• DeepHEM demonstrated superior performance compared to existing domain adaptation methods.
• Predicted essentiality scores showed significant correlation with biological indicators.
• Ablation studies validated the effectiveness of individual framework components.
• Case study confirmed 80% of top predictions with existing literature.

Conclusions:

• DeepHEM offers a robust and effective framework for predicting human miRNA essentiality.
• The multi-modal feature extraction and cross-domain alignment enhance prediction accuracy.
• This approach facilitates a deeper understanding of miRNA roles in biological systems.