TRAC

TRAC predicts and classifies antibiotic resistance genes in genomic and metagenomic sequences using alignment-free neural networks and transfer learning.

Key Features:

• Alignment-Free Approach: Uses neural networks that do not rely on sequence alignments to classify antibiotic resistance genes.
• Transfer Learning: Applies transfer learning to adapt pre-trained models for antibiotic resistance class prediction.
• Curated Multi-Database Dataset: Trained on a curated dataset aggregated from 15 different databases annotated with antibiotic class labels.
• Comparison to Alignment-Based Methods: Demonstrated superior performance relative to alignment-based approaches such as BLAST and HMMER for antibiotic resistance prediction.

Scientific Applications:

• Resistance Gene Identification: Identification and classification of antibiotic resistance genes in genomic and metagenomic samples.
• Mechanism and Epidemiology Studies: Supporting studies of resistance mechanisms and the distribution of resistance classes across samples.
• Clinical and Public Health Research: Informing research on prevention, diagnosis, and treatment strategies for infections caused by resistant bacteria.

Methodology:

Alignment-free neural network models trained via transfer learning on a curated dataset assembled from 15 databases with antibiotic class label annotations.