Uncertainty-Aware Biomarker Discovery for Alzheimer's Disease Reversal: Bridging Mouse Models and Human Translation with Conformal Prediction

Recent evidence that P7C3-A20 reverses advanced Alzheimer’s disease (AD) pathology in aged mice—normalizing 174 differentially expressed proteins and restoring cognition (Chaubey et al., 2026)—raises a critical translational question: which of these biomarker changes will replicate in human patients? We propose Uncertainty-Aware Biomarker Discovery (UABD), a methodological framework that applies conformal prediction to provide distribution-free coverage guarantees for cross-species biomarker prioritization under the covariate-shift assumption. UABD constructs prediction sets for human biomarker responses using mouse proteomics as source data, employing weighted conformal prediction to correct for covariate shift between species. In an illustrative analysis mapping mouse proteomic reversal signatures to published human AD cohort biomarker measurements, UABD achieves empirical marginal coverage (≥90%) while identifying a concordant subset of 8 out of 12 examined AD biomarkers whose disease-associated effect directions align across species with calibrated prediction intervals. We further show that plasma p-tau217—the leading translatable AD biomarker—falls within the conformal prediction set, consistent with its potential utility as a candidate pharmacodynamic endpoint for human trials of NAD+-enhancing therapies. No prospective human treatment-response data are used; the results illustrate the framework’s potential for hypothesis generation and endpoint prioritization.