To Explore Antibiotic Resistance Of Urban Environmental Microorganism And Its Influencing Factor Based On Deep Learning

Urban environment microbiota,which is closely related to the activities of human being,is a major reservoir of antibiotic resistance genes(ARGs).Microbes could tolerate higher concentrations of antibiotics because of the presence of ARGs,that could potentially be transferred to bacterial pathogens and threaten the global public health.Therefore,the accurate identification of ARGs is extremely important for addressing the challenge of antibiotic resistance.However,most of the widely used methods are sequence alignment based,which suffered from the weakness of identifying nonhomologous ARGs.In this paper,we integrated convolutional neural network(CNN)and long and short memory neural network(LSTM)based on deep learning theory,and then proposed an alignment independent approach.This method shows robust performance in the test set with the accuracy reaching to 0.9883,0.8565 for the recall,and 0.9664 for the AUC value.Compared to current popular methods,the accuracy of classifying ARG and the recall are improved by 9.5% and 16.4%,respectively,which result in more non-homologous ARGs being identified.With the new proposed method,457,777 ARGs are identified and grouped into 35 resistance categories from 3,741 metagenome samples collected by The Metagenomics and Metadesign of Subways and Urban Biomes(MetaSUB)International Consortium.More than 80% of the samples harbor the ARGs resistant to beta-lactam,aminoglycoside and bacitracin.To explore the regional specificity of antibiotic resistance,we analyze the proportion,abundance,and diversity of ARG in samples from different countries.The results show that the frequency of ARGs observed significantly vary among different countries.The samples obtained from New Zealand,Nigeria and United States harbor the most abundance and diversity of ARGs,while both the abundance and diversity of ARGs are the lowest in the samples obtained from north Europe.Pairwise permutational multivariate analysis of variance reveals that the resistance category abundance profile can distinguish most of the countries from each other.Moreover,the factor analysis result shows that the abundance of ARGs is significantly associated with the antibiotic consumption,gross domestic product per captia and several other socioeconomic and medical indicators.The association with the human gut resistome is of particular concern in our study and we found that the category abundance profile was significantly positively corelated between the human gut and the urban environment microbial communities,which further indicated the close association of urban environment resistome with the human health.This comprehensive study of environmental resistance in global cities can provide a useful resource for addressing the challenge of antibiotic resistance.