Introduction
Human A(H5N8) vaccine datasets are increasingly important for pandemic preparedness because contemporary A(H5) viruses continue to circulate across animal reservoirs and occasionally infect humans, while directly observed human immunodynamic data remain limited. Sparse sampling complicates interpretation because peak timing, threshold crossing, booster dependence, and cross-reactive breadth must be inferred from only a few post-vaccination observations.
Methods
We performed an open-data quantitative reanalysis to determine whether a compact mechanistically constrained framework could recover interpretable homologous and heterologous humoral structure without overstating predictive performance. We combined physics-informed reconstruction with an explicit Kernel ODE comparator, semipooled subject-level analysis, stability checks, exploratory cellular correlation analysis, and external benchmarking.
Results
Across both MN and HI readouts, the PINN and Kernel ODE agreed on the direction of the principal between-group contrasts in early area under the curve and threshold-crossing day, while semipooled estimates were compatible with lower booster dependence and earlier peak timing in previously vaccinated individuals. Exploratory cellular analysis identified the clearest association between IFNg_H5 and CD4_H5, and an independent external H5 dataset showed coherent cross-reactive structure. However, the external CoronaVac benchmark favored the Kernel ODE, indicating that the contribution of this study is a mechanistically interpretable framework for sparse vaccine datasets rather than a universally superior predictor.
Conclusions
These findings support the cautious use of physics-informed reconstruction to organize sparse human vaccine serology into biologically interpretable response patterns. The proposed framework is most valuable as an interpretable sparse-data inference strategy, rather than as evidence of universal predictive superiority of PINN-based modeling.