ONN4MST

ONN4MST performs ontology-aware deep learning to assign microbial community samples to ecological source niches for microbial source tracking.

Key Features:

• Ontology-Aware Structure: Incorporates a tree-like hierarchical structure that mirrors the taxonomical organization of microbial niches to facilitate nuanced classification.
• Deep Learning Architecture: Employs a six-layer neural network architecture aligned with the ontology's layers to produce hierarchical classification results and enhance interpretability.
• High Accuracy and Speed: Demonstrated accuracy of 0.99 and AUC of 0.97 in experiments involving over 125,823 samples from 114 niches.
• Robustness Across Niches: Handles heterogeneous sample sets across diverse ecological niches, supporting large-scale applications.
• Interpretable Results: Produces hierarchical, interpretable classifications that aid inference of microbial origins, including less-studied niches.

Scientific Applications:

• Microbial Source Tracking: Assigns the ecological origin of microbial samples to support studies of microbial ecology and dynamics.
• Contaminant Detection: Detects microbial contaminants to support quality control and safety assessments in various environments.
• Cross-Niche Analysis: Identifies similar microbial communities across ontologically remote niches to enable comparative ecological analyses.

Methodology:

Implements an ontology-aware deep learning approach using a tree-like hierarchical ontology and a six-layer neural network aligned with the ontology layers to produce hierarchical classifications.