List of participants and affiliations:
- (Team Leader) Marcus Nguyen, ANL (Argonne National Lab)
- (Writer) Nicole Bowers, ANL (Argonne National Lab)
- (Tech Lead) Clark Cucinell, ANL (Argonne National Lab)
- (
) Curtis Hendrickson, UAB (University of Alabama at Birmingham) - ( ) Don Dempsey, UAB (University of Alabama at Birmingham)
- ( ) Andrew Warren, BII (University of Virginia Biocomplexity Institute and Initiative)
The primary goal of this project is to establish robust correlations between the results of Antimicrobial Susceptibility Testing (AST) and the presence of Antimicrobial Resistance (AMR) genes, both plasmid-borne and chromosomal. By systematically analyzing bacterial isolates, we aim to identify specific patterns in resistance profiles that correspond to the presence of particular AMR genes and their location on plasmids or the chromosome. This will provide valuable insights into the mechanisms by which resistance is conferred and transmitted within microbial populations, potentially informing future therapeutic strategies and public health interventions aimed at combating the spread of antimicrobial resistance.
This project will employ a 3-pronged approach to systematically identify AMR genes on both plasmids and chromosomal DNA, and correlate them with antimicrobial resistance phenotypes. Using open-source data, including sequences from NCBI, we will first predict whether genomic sequences or contigs originate from plasmids or chromosomal regions. Next, we will apply AMR gene identification algorithms to detect the presence of resistance genes. Additionally, we augment the AST data using phenotype prediction models, thereby increasing our set of antimicrobial resistance profiles. By integrating plasmid predictions with AMR gene data, we will categorize AMR genes based on their genomic location (plasmid vs. chromosomal). These results will then be correlated with the phenotypic resistance data to assess the relationship between the genetic context of AMR genes and observed antimicrobial susceptibility. This approach will enable a detailed analysis of the genomic architecture of resistance and its phenotypic expression.
- Bioconda link
- Website
- In Silico Detection and Typing of Plasmids using PlasmidFinder and Plasmid Multilocus Sequence Typing
Full breakdown on species and individual antibiotics available in the AMR Phenotypes directory.
Species-antibiotic summary
| Ground Truth | Expected | ||||
|---|---|---|---|---|---|
| Species | F1 Score | Samples | F1 Score | 95% CI | |
| acinetobacter baumannii | 0.8975 | 920 | 0.9328 | 0.9321 | 0.9335 |
| campylobacter jejuni | 0.9948 | 2138 | 0.9849 | 0.9847 | 0.9852 |
| clostridioides difficile | 0.0000 | 1 | 0.9678 | #DIV/0! | #DIV/0! |
| enterococcus faecium | 0.3406 | 14 | 0.9711 | 0.9707 | 0.9716 |
| klebsiella pneumoniae | 0.8497 | 469 | 0.8960 | 0.8952 | 0.8969 |
| neisseria gonorrhoeae | 0.5939 | 458 | 0.9640 | 0.9635 | 0.9645 |
| pseudomonas aeruginosa | 0.8398 | 324 | 0.7894 | 0.7873 | 0.7914 |
| salmonella enterica | 0.9230 | 7032 | 0.9319 | 0.9311 | 0.9327 |
| staphylococcus aureus | 0.8324 | 456 | 0.9764 | 0.9761 | 0.9767 |
| streptococcus pneumoniae | 0.9281 | 132 | 0.9629 | 0.9623 | 0.9634 |
AMR Gene Correlations to Phenotype Data
Average and 95% confidence intervals for genes in different gene sets correlated to all antibiotics' phenotypes.
- ALL: All AMR Genes
- Plasmid: AMR genes that are likely to be on plasmids. Gene appears on a plasmid >X% of the time (X=10, 20)
- Chromosome: AMR genes that are likely to be on chromosomes. Gene appears on a chromosome >X% of the time (X=10,20)
- NDARO: Public NDARO AMR Phenotypes
- Augmented: NDARO AMR phenotypes + augmented AMR phenotype predictions from ML model.
| 10% Threshold | ||||
|---|---|---|---|---|
| Gene Correlation | Dataset | Avg AST Corr | 95% CI | |
| ALL | Augmented | 0.0115 | 0.0006 | 0.0224 |
| ALL | NDARO | 0.0179 | 0.0059 | 0.0299 |
| Plasmid | Augmented | 0.0663 | 0.0476 | 0.0850 |
| Plasmid | NDARO | 0.0668 | 0.0464 | 0.0872 |
| Chromosome | Augmented | -0.0310 | -0.0483 | -0.0136 |
| Chromosome | NDARO | -0.0217 | -0.0409 | -0.0024 |
| 20% Threshold | ||||
|---|---|---|---|---|
| Gene Correlation | Dataset | Avg AST Corr | 95% CI | |
| ALL | Augmented | 0.011484778 | 0.000600566 | 0.02236899 |
| ALL | NDARO | 0.017907133 | 0.005891491 | 0.029922775 |
| Plasmid | Augmented | 0.066931452 | 0.052262277 | 0.081600626 |
| Plasmid | NDARO | 0.073473773 | 0.057341983 | 0.089605563 |
| Chromosome | Augmented | -0.030951402 | -0.048302525 | -0.01360028 |
| Chromosome | NDARO | -0.021679668 | -0.040932581 | -0.002426754 |
Arbitrary thresholds set
- Plasmid thresholds for correlation analysis still needs to be fleshed out?
- Filtered out AMR genes that occured in less than 5% of genomes.
Denovo vs guided assemblies
- We only did predictions on denovo assemblies, we need to figure out how to de-duplicate the guided and denovo if we combine them.
Filtering AMR Genes
- Removed low occurence genes: should we have rolled them into the "super family". In other words, if BLA-8 was low occurence, should it's count be added towards BLA?
- Filtering out gene-antibiotic relations that have relation, this would strengthen the correlations, though probably not the final result of plasmid > all > chromosome
Plasmid prediction
- Run anslysis on PlasmidFinder
- Look into failed plasmid prediction runs
AMR phenotypes
- Stuck to mainly susceptible and resistant, correlation to MICs might offer a better depth of correlation.
- Keep the augmented set? For augmented dataset we do need to check if any of the NDARO data was used in training. We also want to remove the non-representative species (like C. diff and Neisseria) if we go this route.
- We augmented only on denovo assemblies and ignored guided. We would need to figure out a way to do guided.
This software was created as part of an NCBI codeathon, a hackathon-style event focused on rapid innovation. While we encourage you to explore and adapt this code, please be aware that NCBI does not provide ongoing support for it.
For general questions about NCBI software and tools, please visit: NCBI Contact Page