A Bayesian modelling framework to improve antibody titer estimation applied to RSV dilution series data

Accurately measuring antibody levels is important for assessing population immunity and guiding vaccine development. We identify batch-level biases and experimental noise in neutralizing antibody (nAb) titer estimates from respiratory syncytial virus (RSV) foci reduction neutralization tests (FRNTs) when using off-the-shelf methods such as the Kärber formula and four-parameter logistic (4PL) model. To address this, we develop a Bayesian hierarchical model (BHM) to estimate nAb titers, correcting for batch effects and other sources of experimental variation. We evaluate model performance using both simulated and experimental FRNT data. In simulation, nAb titers are most accurate using the BHM (Spearman $\rho$ = 0.96 compared to simulation truth, P < 0.001; root mean square error [RMSE] = 0.41), outperforming the Kärber formula ( $\rho$ = 0.63, P < 0.001; RMSE = 1.64) and 4PL model ( $\rho$ = 0.87, P < 0.001; RMSE = 1.09). The Kärber formula produces more false negatives (9.85%) than the 4PL model (2.42%) and BHM (0.93%), and the 4PL model often produces biased titers for weakly positive samples. In experimental data, population-level measures, such as geometric mean titers (GMTs), seroprevalence and seroincidence differ substantially depending on the method. This framework can be adapted to other antibody assays producing dilution series data and improves the accuracy and robustness of titer estimates across a range of experimental settings.