Estimating time since influenza virus exposure using single-cell proteomic data

Determining when the onset of a respiratory infection occurred is important for effective clinical management and can aid in mapping transmission events. However, current diagnostic assays report only pathogen detection status and do not provide any information about the timing of infection, in part because of the lack of biomarkers that inform time since exposure.

To address this gap, we developed immune-based predictive models of infection timing and shedding status using data from a controlled human challenge with influenza A/California/2009 (H1N1), in which major immune cell subsets were longitudinally profiled across multiple time points before and after viral challenge using 42-marker mass cytometry panels. Random forest machine learning models were trained to address two predictive objectives: (1) distinguishing virus shedders from non-shedders and (2) estimating days post-infection challenge (DPC) from immune profiles. Model performance was evaluated within the primary challenge cohort and independently validated using data from a separate controlled human influenza challenge study using the same virus.

Our analysis revealed that single-cell immune population dynamics alone encode a robust and reproducible temporal structure following influenza infection, enabling accurate estimation of virus exposure timing.

These findings provide foundational insight into host immune responses during influenza infection and represent an early step toward a future class of immune-based diagnostics that could extend beyond pathogen detection to inform infection timing and the duration of periods associated with contagiousness.