deepNEC

deepNEC identifies and classifies enzymes involved in nitrogen biochemical networks using an alignment-free deep learning approach.

Key Features:

• Alignment-Free Approach: Bypasses traditional alignment-based methods by extracting features directly from protein sequences for enzyme prediction.
• Deep Learning Framework: Employs a multilayer deep learning model on encoded protein-sequence features to learn complex patterns associated with enzyme function.
• Comprehensive Feature Extraction: Utilizes amino acid composition, dipeptide composition (DPC), conformation transition and distribution, normalized Moreau-Broto autocorrelation (NMBroto), conjoint and quasi order features.
• Four-Tier Prediction Model: Implements a four-phase pipeline that (Phase I) distinguishes enzymes from non-enzymes, (Phase II) classifies enzymes as nitrogen biochemical network-related or non-nitrogen metabolism, (Phase III) assigns nitrogen-related enzymes into nine specific nitrogen metabolism classes, and (Phase IV) predicts the enzyme commission (EC) number from 20 possible classes for nitrogen metabolism.
• Performance Metrics: Demonstrates accuracy exceeding 93% on independent testing and a Matthews correlation coefficient up to 0.92 during training, with the DPC+NMBroto hybrid feature set yielding the best results across phases.
• Homology-Based Validation: Incorporates a homology-based method to reduce false negatives and improve prediction reliability.
• Input Data: Operates on protein sequence data including enzymatic and non-enzymatic sequences.

Scientific Applications:

• Metagenomics: Facilitates functional annotation of genes encoding nitrogen network enzymes in metagenomic datasets.
• Agriculture: Supports analysis of microbial enzymes that influence soil nitrogen cycling relevant to agricultural research.
• Wastewater Treatment: Aids identification of nitrogen-transforming enzymes within wastewater microbial communities.
• Industrial Biotechnology: Assists discovery and classification of enzymes for nitrogen-related bioprocesses in industrial biotechnology.
• Functional Annotation: Enables functional annotation of genes encoding core nitrogen network enzymes across diverse sequence datasets.

Methodology:

Model training used two datasets of enzymatic and non-enzymatic protein sequences with alignment-free, sequence-derived features fed to a multilayer deep learning model; validation employed k-fold cross-validation and independent testing, and a homology-based method was applied to minimize false negatives.