Antilogic, a new supervised machine learning software for the automatic interpretation of antibiotic susceptibility testing in clinical microbiology: proof-of-concept on three frequently isolated bacterial species.

OBJECTIVE: Antibiotic susceptibility testing (AST) is necessary in order to adjust empirical antibiotic treatment, but the interpretation of results requires experience and knowledge. We have developed a machine learning software that is capable of reading AST images without any human intervention and that automatically interprets the AST, based on a database of antibiograms that have been clinically validated with European Committee on Antimicrobial Susceptibility Testing rules. METHODS: We built a database of antibiograms that were labelled by senior microbiologists for three species: Escherichia coli, Klebsiella pneumoniae, and Staphylococcus aureus. We then developed Antilogic, a Python software based on an original image segmentation module and supervised learning models that we trained against the database. Finally, we blind tested Antilogic against a validation set of 5100 photos of antibiograms. RESULTS: We trained Antilogic against a database of 18072 pictures of antibiograms. Overall agreement against the validation set reached 97% (16 855/17 281) regarding phenotypes. The severity rate of errors was also evaluated: 1.66% (287/17 281) were major errors and 0.80% (136/17 281) were very major errors. After implementation of uncertainty quantifications, the rate of errors decreased to 0.80% (114/13 451) and 0.42% (51/13 451) for major and very major errors respectively. DISCUSSION: Antilogic is the first machine learning software that has been developed for AST interpretation. It is based on a novel approach that differs from the typical diameter measurement and expert system approach. Antilogic is a proof of concept that artificial intelligence can contribute to faster and easier diagnostic methods in the field of clinical microbiology.

Adapted Protocol for Saccharibacteria Cocultivation: Two New Members Join the Club of Candidate Phyla Radiation.

The growing application of metagenomics to different ecological and microbiome niches in recent years has enhanced our knowledge of global microbial biodiversity. Among these abundant and widespread microbes, the candidate phyla radiation (CPR) group has been recognized as representing a large proportion of the microbial kingdom (>26%). CPR are characterized by their obligate symbiotic or exoparasitic activity with other microbial hosts, mainly bacteria. Currently, isolating CPR is still considered challenging for microbiologists. The idea of this study was to develop an adapted protocol for the coculture of CPR with a suitable bacterial host. Based on various sputum samples, we tried to enrich CPR (Saccharibacteria members) and to cocultivate them with pure hosts (Schaalia odontolytica). This protocol was monitored by TaqMan real-time quantitative PCR (qPCR) using a system specific for Saccharibacteria designed in this study, as well as by electron microscopy and sequencing. We succeeded in coculturing and sequencing the complete genomes of two new Saccharibacteria species, "Candidatus Minimicrobia naudis" and "Candidatus Minimicrobia vallesae." In addition, we noticed a decrease in the CT values of Saccharibacteria and a significant multiplication through their physical association with Schaalia odontolytica strains in the enriched medium that we developed. This work may help bridge gaps in the genomic database by providing new CPR members, and in the future, their currently unknown characteristics may be revealed. IMPORTANCE In this study, the first TaqMan real-time quantitative PCR (qPCR) system, targeting Saccharibacteria phylum, has been developed. This technique can specifically quantify Saccharibacteria members in any sample of interest in order to investigate their prevalence. In addition, another easy, specific, and sensitive protocol has been developed to maintain the viability of Saccharibacteria cells in an enriched medium with their bacterial host. The use of this protocol facilitates subsequent studies of the phenotypic characteristics of CPR and their physical interactions with bacterial species, as well as the sequencing of new genomes to improve the current database.

Full-length title: NRPPUR database search and in vitro analysis identify an NRPS-PKS biosynthetic gene cluster with a potential antibiotic effect.

BACKGROUND: Growing concern about the emergence of antibiotic resistance is compelling the pharmaceutical industry to search for new antimicrobial agents. The availability of genome sequences has enabled the development of computational mining as an important tool in the discovery of natural products with antibiotic effect. RESULTS: NRPPUR (Non-Ribosomal Peptide and Polyketide Urmite) is a new bioinformatic tool that was created to detect polyketides and non-ribosomal peptide gene clusters (PKS and NRPS) in bacterial genomes using the rpsBlast program. The NRPPUR database was constructed locally by assembling all 3505 available sequences of NRPS-PKS that have been identified by in silico approaches to date, with 164 Biosynthetic Gene Clusters (BGCs) derived from the published literature that have demonstrated antimicrobial activity in vitro. The in silico analysis of 49 intestinal human bacterial genomes using the NRPPUR made it possible to identify 91 BGCs including 89 clusters that had never previously been described. On average, intestinal human bacterial genomes devote nearly 0.8% (±1.4% s.d.) of their genome to NRPS/PKS biosynthesis, with Bacillus vallismortis, Streptomyces massiliensis and Bacillus subtilis genomes apportioning 8.4, 3.6 and 3.15% of their genomes, respectively. When using the cross-streak method, S. massiliensis displayed antibacterial activity against many Gram-positive and negative bacteria including methicillin-resistant Staphylococcus aureus (MRSA). CONCLUSIONS: NRPPUR has proven to be a very useful tool for the primary in silico selection of species with potential antimicrobial activity and human microbiota could be the future source of new antimicrobial discoveries. Further exploration of this and other ecological niches, coupled with high-throughput antibacterial activity screening should be envisaged.