A phylogenetic method to perform genome-wide association studies in microbes that accounts for population structure and recombination

Collins, Caitlin; Didelot, Xavier

doi:10.1101/140798

Cited by 60 publications

(88 citation statements)

References 79 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The algorithms and their performance guarantees could be adapted to other types of representations for genomic variants, such as single nucleotide polymorphisms (SNP) and unitigs 45 . The techniques proposed by Hardt et al (2016) 46 could be used to ensure that the models are not biased towards undesirable covariates, such as population structure 47,48 . This could potentially increase the interpretability of models, by avoiding the inclusion of rules that are associated with biases in the data.…”

Section: Discussionmentioning

confidence: 99%

Interpretable genotype-to-phenotype classifiers with performance guarantees

Drouin

Letarte

Raymond

et al. 2018

Preprint

View full text Add to dashboard Cite

Understanding the relationship between the genome of a cell and its phenotype is a central problem in precision medicine. Nonetheless, genotype-to-phenotype prediction comes with great challenges for machine learning algorithms that limit their use in this setting. The high dimensionality of the data tends to hinder generalization and challenges the scalability of most learning algorithms. Additionally, most algorithms produce models that are complex and difficult to interpret. We alleviate these limitations by proposing strong performance guarantees, based on sample compression theory, for rule-based learning algorithms that produce highly interpretable models. We show that these guarantees can be leveraged to accelerate learning and improve model interpretability. Our approach is validated through an application to the genomic prediction of antimicrobial resistance, an important public health concern. Highly accurate models were obtained for 12 species and 56 antibiotics, and their interpretation revealed known resistance mechanisms, as well as some potentially new ones. An open-source disk-based implementation that is both memory and computationally efficient is provided with this work. The implementation is turnkey, requires no prior knowledge of machine learning, and is complemented by comprehensive tutorials.

show abstract

Section: Discussionmentioning

confidence: 99%

Interpretable genotype-to-phenotype classifiers with performance guarantees

Drouin

Letarte

Raymond

et al. 2018

Preprint

View full text Add to dashboard Cite

show abstract

“…We chose methods which employ different approaches to correct for population structure. We compared FaST-LMM (Lippert et al 2011), which uses SNP calls to compute kinship, an implementation of Pyseer (Lees et al 2018) which uses a phylogenetic tree to measure relatedness, and treeWAS (Collins and Didelot 2018), which uses the phylogenetic tree to identify genotype-phenotype correlations that are stronger than expected based on shared ancestry and recombination. Due to the computation time required to run treeWAS, it was only applied to gene presence/absence data.…”

Section: Identification Of Resistance Loci With Gwasmentioning

confidence: 99%

“…Association tests were performed using FaST-LMM (v.0.2.32) (Lippert et al 2011), Pyseer (v.1.2.0) (Lees et al 2018), and treeWAS (v.1.0) (Collins and Didelot 2018). For FaST-LMM analyses, a kinship matrix was calculated using SNP calls.…”

Section: Genome-wide Association Studiesmentioning

confidence: 99%

Contrasting approaches to genome-wide association studies impact the detection of resistance mechanisms in Staphylococcus aureus

Wheeler

Reuter

Chewapreecha

et al. 2019

Preprint

View full text Add to dashboard Cite

Rapid detection of antibiotic resistance using whole-genome sequencing (WGS) could improve clinical outcomes and limit the spread of resistance. For this to succeed, we need an accurate way of linking genotype to phenotype, that identifies new resistance mechanisms as they appear. To assess how close we are to this goal, we characterized antimicrobial resistance determinants in >4,000 Staphylococcus aureus genomes of isolates associated with bloodstream infection in the United Kingdom and Ireland. We sought to answer three questions: 1) how well did known resistance mechanisms explain phenotypic resistance in our collection, 2) how many previously identified resistance mechanisms appeared in our collection, and 3) how many of these were detectable using four contrasting genome-wide association study (GWAS) methods. Resistance prediction based on the detection of known resistance determinants was 98.8% accurate. We identified challenges in correcting for population structure, clustering orthologous genes, and identifying causal mechanisms in rare or common phenotypes, which reduced the recovery of known mechanisms. Limited sensitivity and specificity of these methods made prediction using GWAS-discovered hits alone less accurate than using literature-derived genetic determinants. However, GWAS methods identified novel mutations associated with resistance, including five mutations in rpsJ, which improved tetracycline resistance prediction for 28 isolates, and a T118I substitution in fusA which resulted in better fusidic acid resistance prediction for 5 isolates. Thus, GWAS approaches in conjunction with phenotypic testing data can support the development of comprehensive databases to enable real-time use of WGS for patient management.

show abstract

“…The overall problem of relating microbial genotype to phenotype has generally been approached by genome-wide association study (GWAS) methods (11)(12)(13) . Generally these methods take a univariate approach, considering the association between a phenotype and a single variant, then 'scanning' along the whole genome one variant at a time.…”

Section: Introductionmentioning

confidence: 99%

Improved inference and prediction of bacterial genotype-phenotype associations using pangenome-spanning regressions

Lees

Mai

Wheeler

et al. 2019

Preprint

View full text Add to dashboard Cite

Discovery of influential genetic variants and prediction of phenotypes such as antibiotic resistance are becoming routine tasks in bacterial genomics. Genome-wide association study (GWAS) methods can be applied to study bacterial populations, with a particular emphasis on alignment-free approaches, which are necessitated by the more plastic nature of bacterial genomes. Here we advance bacterial GWAS by introducing a computationally scalable joint modeling framework, where genetic variants covering the entire pangenome are compactly represented by unitigs, and the model fitting is achieved using elastic net penalization. In contrast to current leading GWAS approaches, which test each genotype-phenotype association separately for each variant, our joint modelling approach is shown to lead to increased statistical power while maintaining control of the false positive rate. Our inference procedure also delivers an estimate of the narrow-sense heritability, which is gaining considerable interest in studies of bacteria. Using an extensive set of state-of-the-art bacterial population genomic datasets we demonstrate that our approach performs accurate phenotype prediction, comparable to popular machine learning methods, while retaining both interpretability and computational efficiency. We expect that these advances will pave the way for the next generation of high-powered association and prediction studies for an increasing number of bacterial species.

show abstract

A phylogenetic method to perform genome-wide association studies in microbes that accounts for population structure and recombination

Cited by 60 publications

References 79 publications

Interpretable genotype-to-phenotype classifiers with performance guarantees

Interpretable genotype-to-phenotype classifiers with performance guarantees

Contrasting approaches to genome-wide association studies impact the detection of resistance mechanisms in Staphylococcus aureus

Improved inference and prediction of bacterial genotype-phenotype associations using pangenome-spanning regressions

Contact Info

Product

Resources

About