Biffignandi, G. B. | bioRxiv Science

bioRxiv2025

Ciprofloxacin resistance in Klebsiella pneumoniae: phenotype prediction from genotype and global distribution of resistance determinants

BACKGROUNDCiprofloxacin resistant Klebsiella pneumoniae is common or emerging in many geographies, and knowledge of local resistance rates is important for empirical therapy. Whilst there are known K. pneumoniae ciprofloxacin resistance determinants, there is a lack of systematic data on the effect of determinants, alone and in combination, and there are no publicly accessible tools for predicting resistance from whole genome sequence data. METHODSThe KlebNET-GSP AMR Genotype-Phenotype Group aggregated a matched genotype-phenotype dataset of n=12,167 K. pneumoniae species complex (KpSC) isolates from 27 countries between 2001-2021. We developed a rules-based classifier to predict ciprofloxacin resistance by categorizing the number of quinolone resistance determining regions mutations in gyrA and parC, the number of plasmid-mediated quinolone resistance genes, and the presence/absence of aac(6')-Ib-cr (which can acetylate ciprofloxacin). Predictive performance was assessed using the discovery dataset, for which we re-phenotyped discrepant isolates; and validated using externally contributed datasets (n=7,030 KpSC isolates). RESULTSThe rules-based classifier predicted R vs S/I with categorical agreement, sensitivity, and specificity >96%, and major/very major error rates <4%. Performance was similar across diverse KpSC sources (human, animal, other), species, and intra-species lineages. External validation of the classifier yielded overall 93.12% categorical agreement [95% confidence interval (CI), 92.50-93.74%], 8.65% major errors [95% CI, 7.34-9.97%], and 6.20% very major errors [95% CI, 5.51-6.90%]. We implemented the classifier in Kleborate, a command-line tool that is integrated into the Pathogenwatch web platform. Using this to assess the global distribution of ciprofloxacin resistance determinants in KpSC genomes available in Pathogenwatch (n=31,319, from 109 countries between years 2000-2023), we observed a significant positive association between national quinolone consumption rates and predicted ciprofloxacin resistance (R2=0.20, p=0.004). CONCLUSIONS Ciprofloxacin resistance phenotypes can be reasonably predicted from genotypes, which is sufficient for informing surveillance. However, unexplained resistance remains and accuracy is insufficient for clinical applications. We demonstrate the value of aggregating genotype-phenotype data to explore resistance mechanisms and develop predictors, but highlight complexities in combining phenotype data from different assays and standards.

genomics↗

bioRxiv2023

Optimising machine learning prediction of minimum inhibitory concentrations in Klebsiella pneumoniae

Minimum Inhibitory Concentrations (MICs) are the gold standard for quantitatively measuring antibiotic resistance. However, lab-based MIC determination can be time-consuming and suffers from low reproducibility, and interpretation as sensitive or resistant relies on guidelines which change over time. Genome sequencing and machine learning promise to allow in-silico MIC prediction as an alternative approach which overcomes some of these difficulties, albeit the interpretation of MIC is still needed. Nevertheless, precisely how we should handle MIC data when dealing with predictive models remains unclear, since they are measured semi-quantitatively, with varying resolution, and are typically also left- and right-censored within varying ranges. We therefore investigated genome-based prediction of MICs in the pathogen Klebsiella pneumoniae using 4367 genomes with both simulated semi-quantitative traits and real MICs. As we were focused on clinical interpretation, we used interpretable rather than black-box machine learning models, namely, Elastic Net, Random Forests, and linear mixed models. Simulated traits were generated accounting for oligogenic, polygenic, and homoplastic genetic effects with different levels of heritability. Then we assessed how model prediction accuracy was affected when MICs were framed as regression and classification. Our results showed that treating the MICs differently depending on the number of concentration levels of antibiotic available was the most promising learning strategy. Specifically, to optimise both prediction accuracy and inference of the correct causal variants, we recommend considering the MICs as continuous and framing the learning problem as a regression when the number of observed antibiotic concentration levels is large, whereas with a smaller number of concentration levels they should be treated as a categorical variable and the learning problem should be framed as a classification. Our findings also underline how predictive models can be improved when prior biological knowledge is taken into account, due to the varying genetic architecture of each antibiotic resistance trait. Finally, we emphasise that incrementing the population database is pivotal for the future clinical implementation of these models to support routine machine-learning based diagnostics. Data SummaryThe scripts used to run and fit the models can be found at https://github.com/gbatbiff/Kpneu_MIC_prediction. The Illumina sequences from Thorpe et al. are available from the European Nucleotide Archive under accession PRJEB27342. All the other genomes are available on https://www.bv-brc.org/ database. Impact statementKlebsiella pneumoniae is a leading cause of hospital and community acquired infections worldwide, highly contributing to the global burden of antimicrobial resistance (AMR). Ordinary methods to assess antibiotic resistance are not always satisfactory, and may not be effective in terms of costs and delays, so robust methods able to accurately predict AMR are increasingly needed. Genome-based prediction of minimum inhibitory concentrations (MICs) through machine learning methods is a promising tool to assist clinical diagnosis, also offsetting phenotypic MIC discordance between the different culture-based assays. However, benchmarking predictive models against phenotypic data is problematic due to inconsistencies in the way these data are generated and how they should be handled remains unclear. In this work, we focused on genome-based prediction of MIC and evaluated the performance of interpretable machine learning models across different genetic architectures and data encodings. Our workflow highlighted how MICs need to be treated as different types of data depending on the method used to measure them, in particular considering each antibiotic separately. Our findings shed further light on the factors affecting model performance, paving the way to future improvements of antibiotic resistance prediction.

bioinformatics↗

bioRxiv2020

Identification of variable sites in Sars-CoV-2 and their abundance profiles in time

The SARS-CoV-2 pandemic that we are currently experiencing is exerting a massive toll both in human lives and economic impact. One of the challenges we must face is to try to understand if and how different variants of the virus emerge and change their frequency in time. Such information can be extremely valuable as it may indicate shifts in aggressiveness, and it could provide useful information to trace the spread of the virus in the population. In this work we identified and traced over time 7 amino acid variants that are present with high frequency in Italy and Europe, but that were absent or present at very low frequencies during the first stages of the epidemic in China and the initial reports in Europe. The analysis of these variants helps defining 6 phylogenetic clades that are currently spreading throughout the world with changes in frequency that are sometimes very fast and dramatic. In the absence of conclusive data at the time of writing, we discuss whether the spread of the variants may be due to a prominent founder effect or if it indicates an adaptive advantage.

bioinformatics↗