1426. Integrating artificial intelligence and dynamic programming to identify microbial clusters through antimicrobial susceptibility sequence analysis

BACKGROUND: The spread of infectious diseases and antimicrobial resistance poses a significant threat, especially in developing countries where traditional DNA fingerprinting techniques are often not available. We propose a novel approach using dynamic programming algorithms to compare antimicrobial susceptibility sequences and identify clusters of related microbes. METHODS: We selected pathogens from specific infections and tested them with a range of antimicrobials, recording the test results as either sensitive (S), resistant (R), or unknown (U) to each drug. These results were combined into a sequence, or ATB string, for each pathogen (Fig. 1). To compare the sequences, we used dynamic programming (Fig. 2) to calculate the minimum edit distance between each pair of sequences. The edit distance reflects the number of atomic changes needed to transform one sequence into the other. In this case, atomic changes include insertion, deletion, substitution, and matching of individual test results (Fig. 3). We created a matrix of pairwise similarities by comparing all ATB strings. A cluster of similar strains was identified based on a cutoff in the edit distance matrix, such as a distance of less than 1. The method, which is available in the SACIH+ system (https://plus.sacihweb.com), was used to analyze the same strain of specific healthcare infections from a public hospital in Belo Horizonte, Brazil, diagnosed between Jan 2022 and Mar 2023. [Figure: see text] [Figure: see text] RESULTS: We analyzed S. aureus strains from surgical site infection - SSI (23 cases), pneumonia - PNEU (29 cases), and bloodstream infection - BSI (13 cases). Identified 4 SSI, 10 PNEU clusters (Fig. 4 and 5), and 4 BSI clusters using edit distance matrix. For K. pneumoniae strains from 21 BSI, 22 PNEU, and 34 urinary tract infection - UTI cases, found 5 clusters for PNEU and UTI, and 7 clusters for BSI (Fig. 6). [Figure: see text] [Figure: see text] [Figure: see text] CONCLUSION: Our novel approach using dynamic programming to compare antimicrobial susceptibility sequences has shown promise in identifying clusters of related microorganisms, even without traditional DNA fingerprinting methods. The method represents a valuable tool for tracking the spread of infectious diseases, especially in settings where traditional methods are not available. [Figure: see text] DISCLOSURES: All Authors: No reported disclosures