CycleFlow: Learning Transporter Conformational Cycles for Mechanistic Interpretation of Disease Variants

Membrane transporters function through ordered conformational transitions, but experimentally resolved structures usually provide only sparse snapshots of a complete cycle. Existing AI approaches largely treat alternative-structure prediction and variant interpretation as separate problems, leaving a missing mechanistic layer between structural change and disease. We propose CycleFlow, a framework that learns a continuous cycle coordinate, functional transition topology, unresolved intermediate conformations, and transition-specific structural determinants from sparse transporter structures. We further represent a disease-associated variant as a substrate-conditioned perturbation of individual transition steps. The resulting conformational disease signature asks which substrate is affected and at which step of the cycle, providing a falsifiable route toward mechanistic variant interpretation and state-aware therapeutic hypotheses.