prewas: data pre-processing for more informative bacterial GWAS

Saund, Katie; Lapp, Zena; Thiede, Stephanie; Pirani, Ali; Snitkin, Evan S.

doi:10.1099/mgen.0.000368

Cited by 12 publications

(15 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To this end, we independently evaluated patient characteristics as well as three different genomic feature sets for their ability to classify colonization and infection. The three genomic feature sets were uncurated genomic (including SNPs, indels, IS elements, and accessory genes), uncurated grouped genomic (variants grouped into genes, akin to a burden test, e.g (28)), and curated genomic (features identified using Kleborate (16)). Across the 100 different train/test splits, we observed that the average predictive performance was weak, with each of the genomic and patient feature sets predictive of infection to a similar degree (all 1st quartile AUROCs > 0.5; median range=0.55-0.68; Figure 2A ; AUPRC: Figure S3A ).…”

Section: Resultsmentioning

confidence: 99%

Machine learning models to identify patient and microbial genetic factors associated with carbapenem-resistantKlebsiella pneumoniaeinfection

Lapp¹,

Han

Wiens³

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

Background Among patients colonized with carbapenem-resistant Klebsiella pneumoniae (CRKP), only a subset develop clinical infection. While patient characteristics may influence risk for infection, it remains unclear if the genetic background of CRKP strains contributes to this risk. We applied machine learning to quantify the capacity of patient characteristics and microbial genotypes to discriminate infection and colonization, and identified patient and microbial features associated with infection across multiple healthcare facilities. Methods Machine learning models were built using whole-genome sequences and clinical metadata from 331 patients colonized or infected with CRKP across 21 long-term acute care hospitals. To quantify variation in performance, we built models using 100 different train/test splits of the entire dataset, and urinary and respiratory site-specific subsets, and evaluated predictive performance on each test split using the area under the receiver operating characteristics curve (AUROC). Patient and microbial features predictive of infection were identified as those consistently important for predicting infection based on average change in AUROC when included in the model. Findings We found that patient and genomic features were only weakly predictive of clinical CRKP infection vs. colonization (AUROC IQRs: patient=0.59-0.68, genomic=0.55-0.61, combined=0.62-0.68), and that one feature set did not consistently outperform the other (genomic vs. patient p=0.4). Comparable model performances were observed for anatomic site-specific models (combined AUROC IQRs: respiratory=0.61-0.71, urinary=0.54-0.64). Strong genomic predictors of infection included the presence of the ICEKp10 mobile genetic element carrying an iron acquisition system (yersiniabactin) and a toxin (colibactin), along with disruption of an O-antigen biosynthetic gene in a sub-lineage of the epidemic ST258 clone. Teasing apart sequential evolutionary steps in the context of clinical metadata indicated that altered O-antigen biosynthesis increased association with the respiratory tract, and subsequent acquisition of ICEKp10 was associated with increased virulence. Interpretation Our results support the need for rigorous machine learning frameworks to gain realistic estimates of the performance of clinical models of infection. Moreover, integrating microbial genomic and clinical data using such a framework can help tease apart the contribution of microbial genetic variation to clinical outcomes. Funding Centers for Disease Control and Prevention, National Institutes of Health, National Science Foundation

show abstract

Section: Resultsmentioning

confidence: 99%

Machine learning models to identify patient and microbial genetic factors associated with carbapenem-resistantKlebsiella pneumoniaeinfection

Lapp¹,

Han

Wiens³

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…We preprocessed variants to include multiallelic sites and used the major allele method for variant binarization, as described in Saund et al 35 We used SnpEff to predict the functional impact of single nucleotide variants and indels (high, moderate, low, modifier). 36 Additionally, we considered all insertions in or upstream of genes as high impact, and those downstream of genes as moderate impact.…”

Section: Variant Preprocessingmentioning

confidence: 99%

Fitness barriers to spread of colistin resistance overcome by first establishing niche in patients with enhanced colistin exposure

Lapp

Han

Choudhary

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

To combat antibiotic resistance, it is critical to improve our understanding of how new resistant strains emerge and spread. An antibiotic resistance threat of critical priority is the epidemic ST258 strain of carbapenem-resistant Klebsiella pneumoniae (CRKP). Here, we studied the spread of resistance among ST258 to an antibiotic of last resort, colistin, by tracking its evolution across 21 U.S. long-term acute care hospitals over the course of a year. Phylogenetic analysis suggested that a significant cost was associated with colistin resistance in most cases, as resistance emergence was common but resistance variants were rarely transmitted. The high cost of resistance was further supported by the observation that several of the resistance variants that were transmitted had acquired secondary variants that reverted the strain to colistin susceptibility. The exceptions to the general pattern of instability associated with colistin resistance were two large clusters of resistant strains in one ST258 clade II sublineage (clade IIB) present across 11 of the 12 sampled Southern California hospitals. Quantification of transmission fitness in the healthcare environment using time-scaled haplotypic density indicated that while resistant isolates from other clades were less fit than their susceptible counterparts, clade IIB resistant isolates were more fit. Overlaying patient clinical data suggested that the increased fitness of colistin-resistant clade IIB isolates is in part driven by a lineage defining variant that increased clade IIB’s association with patient subpopulations who were more likely to be treated with colistin. These results show that a favorable genetic background and sustained selective pressure led to the emergence and spread of a colistin-resistant ST258 sublineage across a regional healthcare network. More broadly, these findings highlight the utility of integrating pathogen genomic and corresponding clinical data from regional healthcare networks to detect and understand the origin and dissemination of antibiotic resistance threats.

show abstract

“…An optional argument may be supplied to facilitate grouping genotypes. The genotype matrix and tree can be prepared from a multiVCF file by the variant preprocessing tool prewas(15). Hogwash assumes that the genotype is encoded such that 0 refers to wild type and 1 refers to a mutation and that binary phenotypes are encoded such that 0 refers to absence and 1 refers to presence.…”

Section: Package Descriptionmentioning

confidence: 99%

“…Users may supply hogwash with data that was previously grouped (for example, using the group SNPs by gene functionality in prewas(15)) but this approach may mask some genotype transitions. In this case, the user does not need to provide a key and the hogwash grouping step is skipped.…”

Section: Package Descriptionmentioning

confidence: 99%

hogwash: Three Methods for Genome-Wide Association Studies in Bacteria

Saund

Snitkin

2020

Preprint

Self Cite

View full text Add to dashboard Cite

17Bacterial genome-wide association studies (bGWAS) capture associations between genomic 18 variation and phenotypic variation. Convergence based bGWAS methods identify genomic 19 mutations that arise more often in the presence of phenotypic variation than is expected by 20 chance. This work introduces hogwash, an open source R package that implements three 21 algorithms for convergence based bGWAS. Hogwash additionally contains a novel grouping tool 22 to perform gene-or pathway-analysis to improve power and increase convergence detection for 23 related but weakly penetrant genotypes. To identify optimal use cases we applied hogwash to 24 data simulated with a variety of phylogenetic signals and convergence distributions. These 25 simulated data are publicly available and contain the relevant metadata regarding convergence 26 and phylogenetic signal for each phenotype and genotype. Hogwash is available for download 27 from GitHub. 28 29 DATA SUMMARY 30 1. hogwash is available from GitHub under the MIT license 31 (https://github.com/katiesaund/hogwash) and can be installed using the R command 32 devtools::install_github("katiesaund/hogwash") 33 2. The simulated data used in this manuscript and the code to generate it are available 34 from GitHub 35 (https://github.com/katiesaund/simulate_data_for_convergence_based_bGWAS) 36 37 IMPACT STATEMENT 38We introduce hogwash, an R package with three methods for bacterial genome-wide 39 association studies. There are two methods for handling binary phenotypes, including an 40 implementation of PhyC(1), as well as one method for handling continuous phenotypes. We 41 formulate two novel indices quantifying the relationship between phenotype convergence and 42 genotype convergence on a phylogenetic tree, one for binary phenotypes and one for 43 continuous phenotypes. These indices shape an intuitive understanding for the ability of 44 hogwash to detect significant intersections of phenotype convergence and genotype 45 convergence and how to interpret hogwash outputs.46 47 INTRODUCTION 48 Bacterial Genome-Wide Association Studies 49 Bacterial genome-wide association studies (bGWAS) infer statistical associations between 50 genotypes and phenotypes. Seminal bGWAS papers identified novel variants associated with 51 antibiotic resistance in M. tuberculosis and host specificity in Campylobacter(1,2). Since then, 52there have been numerous applications of bGWAS that have further highlighted the potential of 53 this approach to identify genetic pathways underlying phenotypic variation and provide insights 54 into the evolution of phenotypes of interest. Association studies can use various genetic data 55 types including single nucleotide polymorphisms (SNPs), k-mers, copy number variants, 56 accessory genes, insertions, and deletions. To improve the power and interpretability of bGWAS 57 inclusion criteria or weighting can be applied to these variants based on predicted functional 58 impact, membership in pathways of interest, or other user preferences (3,4). Differences 59 between ...

show abstract

prewas: data pre-processing for more informative bacterial GWAS

Cited by 12 publications

References 41 publications

Machine learning models to identify patient and microbial genetic factors associated with carbapenem-resistantKlebsiella pneumoniaeinfection

Machine learning models to identify patient and microbial genetic factors associated with carbapenem-resistantKlebsiella pneumoniaeinfection

Fitness barriers to spread of colistin resistance overcome by first establishing niche in patients with enhanced colistin exposure

hogwash: Three Methods for Genome-Wide Association Studies in Bacteria

Contact Info

Product

Resources

About