A deep ensemble framework for human essential gene prediction by integrating multi-omics data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Essential genes are crucial for organism survival and reproduction.
• Understanding gene essentiality advances basic life science, human disease research, and drug discovery.

Purpose of the Study:

• To propose DeEPsnap, a novel deep neural network method for predicting human essential genes.
• To integrate diverse biological data for enhanced prediction accuracy.

Main Methods:

• DeEPsnap utilizes a snapshot ensemble deep neural network.
• It integrates features from DNA/protein sequences and functional data (gene ontology, protein complexes, domains, interaction networks).
• Cost-sensitive deep neural networks are trained with over 200 integrated features.

Main Results:

• DeEPsnap achieved high predictive performance: 96.16% AUROC, 93.83% AUPRC, and 92.36% accuracy via 10-fold cross-validation.
• Comparative experiments demonstrated DeEPsnap's superiority over traditional and other deep learning models.
• The method effectively predicts human gene essentiality.

Conclusions:

• DeEPsnap is an effective and accurate method for predicting human essential genes.
• The integration of multiple feature types significantly improves prediction.
• This tool has potential applications in understanding fundamental biology and developing therapeutics.