BacT-Seq, a Nanopore-Based Whole-Genome Sequencing Workflow Prototype for Rapid and Accurate Pathogen Identification and Resistance Prediction from Positive Blood Cultures: A Feasibility Study

<b>Background/Objectives</b>: Rapid and accurate pathogen identification and antimicrobial susceptibility testing (AST) are critical for the proper management of patients with bloodstream infection (BSI). Real-time whole-genome sequencing (WGS) represents an attractive opportunity for exhaustive pathogen identification and antimicrobial susceptibility prediction (ASP). This feasibility study introduces BacT-Seq, a WGS-based prototype assay for the rapid and accurate identification of pathogens and the prediction of antimicrobial susceptibility from positive blood cultures using Oxford Nanopore Technologies (ONT) sequencing. <b>Methods</b>: A total of 200 positive blood culture samples from patients with a confirmed BSI were included in this study. DNA isolation from blood cultures was optimized prior to GridION (ONT) sequencing. Pathogen identification and several ASP methods were compared to conventional identification and phenotypic AST methods. <b>Results</b>: Most of the mono-microbial (89%) and poly-microbial (88%) samples were identified by BacT-Seq in less than 10 min of sequencing. While identification of poly-microbial samples remains challenging, identification of mono-microbial samples by sequencing was non-inferior to that of the conventional approach, even revealing an added value in terms of exhaustivity and/or taxonomic resolution. Machine-learning-based ASP models yielded 80% predictions in 2.5 h of sequencing. Their ability to predict resistance phenotypes varied with the microbial species evaluated, from 55/57 (96.5%) for <i>Escherichia coli</i> to 24/48 (50.0%) for <i>Pseudomonas aeruginosa</i>. <b>Conclusions</b>: This study demonstrates the feasibility of implementation of the BacT-Seq platform for the fast and accurate identification of pathogens from positive blood cultures. BacT-Seq performance of resistance predictions by bioinformatics tools is promising but requires further optimization and validation before clinical implementation.