Publicação
Validation of resistome signatures through the application of a machine learning prediction algorithm on metagenomic data
| Resumo: | ABSTRACT- Metagenomic data has been increasingly used in antimicrobial resistance (AMR) studies, but there is still a need for accurate and reliable methods for predicting the relative attribution of AMR determinants to different animal reservoirs. AMR data availability has increased exponentially over the past few years, as has global awareness of the threat that AMR poses to public health, often known as the silent pandemic. This has led to an upsurge in interest in applying machine learning to AMR data. In this study, shot-gun sequences were used from fecal samples of pigs, broilers, turkeys, and veal calves, previously collected during national cross-sectional studies across Europe. The data used in this study corresponded to these samples and their associated relative abundance of AMR determinants. A random forest (RF) model was developed to investigate the relative attribution of AMR determinants to those different reservoirs. Additionally, a descriptive analysis was made to further investigate the 15 most important variables for the RF model. A principal component analysis (PCA) and all-subsets regression were performed to identify reservoir-specific AMR determinants. Ultimately, the reservoir-specific AMR determinants identified here were compared with the resistome signatures identified in a previous study. The results demonstrated that the RF model successfully classified resistomes into corresponding reservoir classes, with high accuracy and reliability. The RF model had more difficulty differentiating pig from veal and broiler from turkey, indicating the similarity of resistome composition between each of these two species. The analyses validated several AMR determinants as resistome signatures of specific animal reservoirs, such as tet(40) and sul2 of veal, tet(Q), mef(A) and cfxA2 of veal and pig, blaTEM-126 of broiler, and tet(A) of broiler and turkey. This study describes a reliable and accurate method for the relative attribution of AMR determinants to different animal reservoirs using metagenomic data. Such results are essential for effective surveillance and control of AMR in animal and human populations |
|---|---|
| Autores principais: | Salgueiro, Helena Sofia Fernandes |
| Assunto: | Random forest Machine learning Metagenómica Resistência aos antimicrobianos Random forest Machine learning Metagenomics Antimicrobial resistance |
| Ano: | 2023 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório da Universidade de Lisboa |
| Resumo: | ABSTRACT- Metagenomic data has been increasingly used in antimicrobial resistance (AMR) studies, but there is still a need for accurate and reliable methods for predicting the relative attribution of AMR determinants to different animal reservoirs. AMR data availability has increased exponentially over the past few years, as has global awareness of the threat that AMR poses to public health, often known as the silent pandemic. This has led to an upsurge in interest in applying machine learning to AMR data. In this study, shot-gun sequences were used from fecal samples of pigs, broilers, turkeys, and veal calves, previously collected during national cross-sectional studies across Europe. The data used in this study corresponded to these samples and their associated relative abundance of AMR determinants. A random forest (RF) model was developed to investigate the relative attribution of AMR determinants to those different reservoirs. Additionally, a descriptive analysis was made to further investigate the 15 most important variables for the RF model. A principal component analysis (PCA) and all-subsets regression were performed to identify reservoir-specific AMR determinants. Ultimately, the reservoir-specific AMR determinants identified here were compared with the resistome signatures identified in a previous study. The results demonstrated that the RF model successfully classified resistomes into corresponding reservoir classes, with high accuracy and reliability. The RF model had more difficulty differentiating pig from veal and broiler from turkey, indicating the similarity of resistome composition between each of these two species. The analyses validated several AMR determinants as resistome signatures of specific animal reservoirs, such as tet(40) and sul2 of veal, tet(Q), mef(A) and cfxA2 of veal and pig, blaTEM-126 of broiler, and tet(A) of broiler and turkey. This study describes a reliable and accurate method for the relative attribution of AMR determinants to different animal reservoirs using metagenomic data. Such results are essential for effective surveillance and control of AMR in animal and human populations |
|---|