DeeplyEssential

DeeplyEssential predicts essential genes in bacteria from gene and protein primary sequences using a deep neural network to identify genes indispensable for survival and inform antimicrobial target discovery.

Key Features:

• Minimal Assumptions on Input Data: Relies solely on gene and protein primary sequences, eliminating dependence on structural or topological features.
• Sequence-only Neural Architecture: Implements a deep neural network architecture tailored to extract predictive signals directly from sequence information.
• Bias Mitigation: Identifies and mitigates a hidden performance bias that affects classifiers on imbalanced datasets and when multiple copies of orthologous genes are present.
• Avoidance of Down-sampling and Clustering: Maintains predictive performance without using down-sampling to balance training sets or clustering to exclude multiple gene copies.
• Superior Performance: Demonstrated improved predictive performance relative to existing classifiers in testing.

Scientific Applications:

• Microbial Genomics: Prediction of essential genes across bacterial genomes using sequence data for comparative and functional genomics studies.
• Antibiotic Target Discovery: Prioritization of genes indispensable for bacterial survival as candidates for antibiotic and antimicrobial agent development.

Methodology:

Training of a deep neural network on sequence data from 30 bacterial species collected from DEG (Database of Essential Genes), with the network learning patterns of gene essentiality directly from gene and protein primary sequences.