From thresholds to trajectories: a perspective on reframing alloimmune risk for computational modeling in solid organ transplantation.

Immunological risk prediction in solid organ transplantation has long depended on threshold-based representation of histocompatibility data: donor-specific antibodies (DSA) reported as positive or negative, mean fluorescence intensity (MFI) assessed by fixed cutoffs, molecular mismatch assigned at transplantation, and assays interpreted at a single time point. Although practical, these conventions simplify the complex and dynamic nature of the immune response. In this perspective, we argue that as machine learning (ML) algorithms and computational approaches enter the field of transplant immunology, we need to focus on whether the inputs that feed into these tools and models reflect how alloimmunity actually behaves. We propose treating alloimmune risk as a time-indexed alloimmune state, updated whenever new data are available, across four domains: antibody profile, molecular mismatch and predicted immunogenicity, recipient immune context, and graft context. Within this framework, the features that experienced clinicians and histocompatibility experts already track (e.g., DSA velocity, persistence, epitope spreading, concordance with graft injury) become computable rather than implicit. We discuss how moving from thresholds to trajectories impacts model design, why HLA laboratory expertise becomes more important rather than less, and why interpretability, regulation, and external validation should precede clinical adoption.