Tacconelli, E. | Search bioRxiv

bioRxiv2026

Linking Host Covariates to COVID-19 Vaccine-Induced Antibody Dynamics in a Large Healthcare Worker Cohort

Antibody responses to COVID-19 vaccination vary widely across individuals, but the factors shaping these differences and their relation to underlying host-response processes are still not fully understood. Here, we combined longitudinal serological data from 4,553 healthcare workers with a nonlinear mixed-effects model to quantify determinants of vaccine-induced antibody dynamics. This framework captured both population-level trends and inter-individual variability in anti-spike antibody trajectories after vaccination. We found that age, sex, and pre-vaccine infection significantly influenced response dynamics. Male sex and increasing age were associated with weaker initial antibody production but slower waning, while individuals with a pre-vaccine infection had pre-existing anti-spike antibodies and slower initial antibody production and faster waning. These effects translated into substantial long-term differences in predicted antibody levels, reaching up to a 27% decrease in antibody levels one year after the third vaccination. Our results provide a dynamic view of the determinants of vaccine-induced humoral immunity and establish a framework for analyzing heterogeneous immune responses in large longitudinal cohorts.

Systems Biology↗

bioRxiv2022

Accessibility of covariance information creates vulnerability in Federated Learning frameworks

Federated Learning (FL) is gaining traction in various fields as it enables integrative data analysis without sharing sensitive data, such as in healthcare. However, the risk of data leakage caused by malicious attacks must be considered. In this study, we introduce a novel attack algorithm that relies on being able to compute sample means, sample covariances, and construct known linearly independent vectors on the data owner side. We show that these basic functionalities, which are available in several established FL frameworks, are sufficient to reconstruct privacy-protected data. Additionally, the attack algorithm is robust to defense strategies that involve adding random noise. We demonstrate the limitations of existing frameworks and propose potential defense strategies analyzing the implications of using differential privacy. The novel insights presented in this study will aid in the improvement of FL frameworks.

bioinformatics↗

bioRxiv2019

Quantifying antibiotic impact on within-host dynamics of extended-spectrum beta-lactamase resistance in hospitalized patients

Antibiotic exposure can perturb the human gut microbiome and cause changes in the within-host abundance of the genetic determinants of drug-resistance in bacteria. Such within-host dynamics are expected to play an important role in mediating the relationship between antibiotic use and persistence of drug-resistance within a host and its prevalence within a population. Developing a quantitative representation of these within-host dynamics is an important step towards a detailed mechanistic understanding of the population-level processes by which antibiotics select for resistance. Here we study extended-spectrum beta-lactamase (ESBL) producing organisms of the Enterobacteriaceae bacterial family. These have been identified as a global public health priority and are resistant to most first-line antibiotics for treatment of Enterobacteriaceae infections. We analyse data from 833 rectal swabs from a prospective longitudinal study in three European countries including 133 ESBL-positive hospitalised patients. Quantitative polymerase chain reaction was used to quantify the abundance of the CTX-M gene family - the most wide-spread ESBL gene family - and the 16S rRNA gene as a proxy for bacterial load. We find strong dynamic heterogeneity in CTX-M abundance that is largely explained by the variable nature of the swab sampling. Using information on time-varying antibiotic treatments, we develop a dynamic Bayesian model to decompose the serial data into observational variation and ecological signal and to quantify the potentially causal antibiotic effects. We find an association of treatment with cefuroxime or ceftriaxone with increased CTX-M abundance (approximately 21% and 10% daily increase, respectively), while treatment with meropenem or piperacillin-tazobactam is associated with decreased CTX-M (approximately 8% daily decrease for both). Despite a potential risk for indirect selection, oral ciprofloxacin is also associated with decreasing CTX-M (approximately 8% decrease per day). Using our dynamic model to make forward stochastic simulations of CTX-M dynamics, we generate testable predictions about antibiotic impacts on duration of carriage. We find that a typical course of cefuroxime or ceftriaxone is expected to more than double a patients carriage duration of CTX-M. A typical course of piperacillin-tazobactam or of meropenem - both options to treat hospital acquired infections (HAI) like pneumonia - would reduce CTX-M carriage time relative to ceftriaxone plus amikacin (also an option to treat HAIs) by about 70%. While most antibiotics showed little association with changes in total bacterial abundance, meropenem and piperacillin-tazobactam were associated with decrease in 16S rRNA abundance (3% and 4% daily decrease, respectively). Our study quantifies antibiotic impacts on within-host resistance abundance and resistance carriage, and informs our understanding of how changes in patterns of antibiotic use will affect the prevalence of resistance. This work also provides an analytical framework that can be used more generally to quantify the antibiotic treatment effects on within-host dynamics of determinants of antibiotic resistance using clinical data.

epidemiology↗