Multimodal data integration to determine viral and innate immune kinetics in human airway epithelium
Abstract
Understanding the mechanisms that govern viral spread in human airway epithelium (HAE) remains a major challenge, particularly with regard to identifying and quantifying key factors such as cell type-specific infectivity, viral transmission paths, and the innate immune dynamics. Although mathematical models and experimental advances have provided valuable insights into respiratory infections, revealing the complex spatio-temporal interactions of infection and immune processes on a tissue-level have remained elusive. Here, we present a novel workflow that combines time-resolved bulk measurements and spatially-explicit image information to allow the inference of viral and immune kinetics within HAE for respiratory viruses. While standard inference methods typically require custom summary statistics and resourceful fitting procedures for each individual data set, our workflow relies on the combination of different, simulation-based trained neural networks using BayesFlow , a framework for neural posterior estimation that allows for amortized inference and the integrative analysis of multimodal data. We validated our approach by simulating viral infection dynamics in HAE using systems of increasing complexity that account for tissue heterogeneity, cell type-specific kinetics and interferon-mediated immune responses, mirroring experimental measurements. Thereby, we could show that integrating spatial information is essential to reliably infer viral transmission kinetics and innate immune interactions on a tissue-level. Applying our approach to experimental data on SARS-CoV-2 infection dynamics within HAE culture systems, we estimated that 84% [58%,100%] of all infections were due to cell-associated transmission, pointing towards local transmission as the dominant mode of SARS-CoV-2 spread within HAE. Our workflow can be readily applied to HAE culture systems for inference of viral and innate immune kinetics of different respiratory viruses, allowing multimodal data integration without the need for frequent resourceful re-fitting approaches.
Related
- An integration-free CRISPR platform for genetic dissection of respiratory virus host dependencies in human nasal epithelium
- Integration of Nascent Transcriptomes and Cross-Species Atlases Reveals Evolutionary Barriers in Transcriptional Kinetics During Mammalian Hibernation.
- Enhancement of gut-microbiome-based classification of prodromal Parkinson's Disease cases by hyperparameter tuning, integration of multimodal data, and meta-analysis.
- HESTIA: Scalable Multimodal Integration of Histology and High-Resolution Spatial Transcriptomics for Robust Spatial Domain Identification
- Engineering developmental transitions during human neurulation through modular integration of multimodal spatial cues
BibTeX
@article{lukas2026multimodal,
title = {{Multimodal data integration to determine viral and innate immune kinetics in human airway epithelium}},
author = {Pascal Lukas and Aurélien Gibeaud and Clarisse Schumer and Jonas Arruda and Jeremie Guedj and Olivier Terrier and Frederik Graw},
journal = {PLoS Computational Biology},
year = {2026},
month = may,
doi = {10.1371/journal.pcbi.1014248},
url = {https://hal.science/hal-05775362v1},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Multimodal data integration to determine viral and innate immune kinetics in human airway epithelium
AU - Pascal Lukas
AU - Aurélien Gibeaud
AU - Clarisse Schumer
AU - Jonas Arruda
AU - Jeremie Guedj
AU - Olivier Terrier
AU - Frederik Graw
PY - 2026
DA - 2026/05/20
JO - PLoS Computational Biology
DO - 10.1371/journal.pcbi.1014248
UR - https://hal.science/hal-05775362v1
AB - Understanding the mechanisms that govern viral spread in human airway epithelium (HAE) remains a major challenge, particularly with regard to identifying and quantifying key factors such as cell type-specific infectivity, viral transmission paths, and the innate immune dynamics. Although mathematical models and experimental advances have provided valuable insights into respiratory infections, revealing the complex spatio-temporal interactions of infection and immune processes on a tissue-level have remained elusive. Here, we present a novel workflow that combines time-resolved bulk measurements and spatially-explicit image information to allow the inference of viral and immune kinetics within HAE for respiratory viruses. While standard inference methods typically require custom summary statistics and resourceful fitting procedures for each individual data set, our workflow relies on the combination of different, simulation-based trained neural networks using BayesFlow , a framework for neural posterior estimation that allows for amortized inference and the integrative analysis of multimodal data. We validated our approach by simulating viral infection dynamics in HAE using systems of increasing complexity that account for tissue heterogeneity, cell type-specific kinetics and interferon-mediated immune responses, mirroring experimental measurements. Thereby, we could show that integrating spatial information is essential to reliably infer viral transmission kinetics and innate immune interactions on a tissue-level. Applying our approach to experimental data on SARS-CoV-2 infection dynamics within HAE culture systems, we estimated that 84% [58%,100%] of all infections were due to cell-associated transmission, pointing towards local transmission as the dominant mode of SARS-CoV-2 spread within HAE. Our workflow can be readily applied to HAE culture systems for inference of viral and innate immune kinetics of different respiratory viruses, allowing multimodal data integration without the need for frequent resourceful re-fitting approaches.
SN - 1553-734X
SN - 1553-7358
ER -