DentaScope-AI: A Real-Time k-mer–Driven CNN-Attention Framework for Rapid Pathogen Identification and Virulence Prediction in Endodontic Infections Using Nanopore Sequencing

Abstract

Acute endodontic infections, including pulp abscess and periapical abscess, are rapidly progressing polymicrobial conditions in which delayed pathogen identification often leads to empirical antibiotic overuse and suboptimal clinical outcomes. Although real-time long-read sequencing offers new opportunities for point-of-care molecular diagnostics, translating high-error Nanopore data into actionable clinical information remains challenging. To address this gap, we propose DentaScope-AI, a real-time k-mer–driven CNN-Attention framework for rapid pathogen identification and virulence prediction in endodontic infections using Nanopore sequencing. The system integrates multi-scale k-mer encoding (k = 5, 7, 9) with a multi-branch one-dimensional convolutional neural network and an attention-based feature fusion module to capture both taxonomic and functional genomic signatures directly from streaming reads. A multi-task learning strategy simultaneously performs species-level classification and multi-label prediction of antibiotic resistance genes and virulence factors, enabling precision antimicrobial guidance. In a curated dataset comprising 2.3 million simulated and clinical Nanopore reads from 25 common oral pathogens, DentaScope-AI achieved 96.4% species-level accuracy, a 93.1% F1-score for antibiotic resistance prediction, and a 91.8% F1-score for virulence factor identification, while generating stable diagnostic outputs within 25 minutes of sequencing initiation. These results highlight the robustness and real-time clinical applicability of the proposed k-mer–driven AI diagnostic framework.