Abstract
Protein-protein interactions governed by conformationally heterogeneous domains remain difficult to drug because ligand-competent states are often absent from single static structures. Here, we present AtlasNMR, a statistical framework that transforms multi-model NMR ensembles into screening-ready conformational hypotheses for small molecule discovery. Using the neuronal nitric oxide synthase (nNOS) PDZ domain that engages the adaptor protein CAPON (NOS1AP) as a model system, AtlasNMR identified two representative conformational states capturing the dominant and minor populations of the NMR ensemble. Ensemble-based virtual screening followed by consensus ranking yielded MC-3 , a small molecule modulator that disrupts the NOS1-NOS1AP interaction in live cells and directly engages the nNOS PDZ domain. MC-3 produced convergent neuroprotective effects in disease-relevant neuronal models by reducing amyloid-β-induced cytotoxicity, suppressing NMDA-driven nitrosative stress, and attenuating pathological tau phosphorylation, while exhibiting a balanced early lead-like ADME and safety profile. Together, this work establishes a generalizable strategy for exploiting NMR ensemble heterogeneity to enable small molecule discovery against dynamic protein-protein interfaces.