Pan-Resistome Analysis And Prediction Of Antibiotic Resistance Phenotypes Of Salmonella

Development of whole-genome sequencing technology and establishment of public databases regarding bacterial antibiotic resistance information have resulted in the building up of software tools for identification of antibiotic resistance characteristics.To our best knowledge,existing tools lack the ability to reveal the diverse distribution of antibiotic resistance genes caused by vertical and horizontal gene transfer.Therefore,based on the concept of“pan-genome”,we analogously proposed“pan-resistome”,which represents the collection of all antibiotic resistance genes within a species and is divided into core and accessory parts.The corresponding tools were developed and utilized to analyze the distribution features of antibiotic resistance genes in Salmonella,a typical foodborne pathogen.In addition,prediction models of antibiotic resistance phenotypes of Salmonella were constructed based on pan-genome and pan-resistome,aiming at providing innovative ideas and practical tools for reveal on the formation and transmission mechanism of bacterial antibiotic resistance.The results of this study are as follows.A software tool PRAP(pan-resistome analysis pipeline)was developed to characterize pan-resistome.PRAP integrated the CARD and Res Finder database,and supported FASTQ,FASTA and Gen Bank format files as input files.This tool has ability to identify antibiotic resistance genes via BLAST or k-mer algorithms and output annotation results,which is followed by implementation of three analysis models:1)Characterize the pan-resistome of input strains and draw plots for distribution of antibiotics resistance genes and pan-resistome curves.2)Classify antibiotic resistance genes and show their distribution among individual strains.3)Describe the distribution of resistance genes related to each antibiotic and predict their individual contribution to the antibiotic resistance phenotypes provided by users.PRAP was used to analyze 16365 Salmonella strains from the Entero Base database to characterize their pan-resistome.Results showed that all antibiotic resistance genes were categorized into 104 alleles,among which 18 alleles belonged to the core resistome involving intrinsic genes that primarily conferred resistance through target alteration and efflux,and86 alleles belonged to the accessory resistome involving acquired genes that conferred resistance through complex mechanisms.Clustering analysis of resistome profiles showed that different evolution branches belonged to various serotypes or STs,and significant differences were observed in the number of antibiotic resistance genes among 12 serotypes and among 11STs,respectively.The proportion of acquired resistance genes and their subtypes revealed that 10 of 23 acquired resistance genes increased yearly,and 73.91%(17/23)of them had only one dominant type.In order to mine the relationship between genomic features and antibiotic resistance phenotypes,based on the genome sequences of 6394Salmonella strains and antimicrobial susceptibility testing results of 15antibiotics obtained from the PATRIC database,we established a method to predict antibiotics resistance phenotypes through whole-genome sequences.The comparison of genomic feature extraction methods demonstrated that pan-genome and pan-resistome methods were more suitable for large-scale machine learning than k-mer method.The comparison of various models showed that the extreme gradient boosting algorithm performed best,and the L₉(3⁴)orthogonal experiment indicated that hyperparameters had no significant effect on the model performance.No significant difference was observed in the performance between models based on pan-genome and pan-resistome,and the high-ranked features extracted from pan-genome were found related to known antibiotic resistant genes.Through the data processing and model optimization,SASP(Salmonella antimicrobial susceptibility prediction tool)was developed based on the models with accuracy reaching 98.62%and 82.05%for predicting antibiotic resistance phenotypes and minimum inhibitory concentrations,respectively.In conclusion,we developed PRAP and SASP,which were successfully used in the rapid identification of antibiotic resistance characteristics of Salmonella.These two software tools are also expected to be widely used in the analysis of antibiotic resistance genes and prediction of antibiotic resistance phenotypes in other bacteria,thus providing referable strategies and schemes for a comprehensive and systematic study on the formation and transmission mechanisms of bacterial antibiotic resistance.