Informed, scalable and unified transport couplings for single-cell alignment and downstream inference with scBIOT

Abstract

Single-cell genomics requires methods that account for technical variation while preserving biological signals, remain tractable at atlas scale and support analyses beyond visualization, but existing approaches discard the fitted correspondence that could connect these tasks. Here we introduce scBIOT, a biologically informed optimal-transport framework that formulates single-cell alignment as an explicit and inspectable coupling between cells. scBIOT incorporates prior biological knowledge to preserve known cell identities, compresses transport through prototypes for large datasets and quantifies gene-level changes induced by alignment using coupling-derived transport scores. Across transcriptomic and spatial mapping benchmarks, scBIOT balanced biological conservation, batch correction and label-transfer accuracy while distinguishing batch-associated stress responses from biological immune and drug-response programs. Prototype transport scaled coupling-based analysis to approximately 100 million perturbed cells. Because the coupling encodes probabilistic correspondences rather than only corrected coordinates, the same fitted coupling supports state mapping, trajectory inference, perturbation analysis, multimodal comparison and spatiotemporal localization. Applications to cancer, hematopoiesis and embryonic development identified condition-specific cell states, modality-dependent regulatory programs and spatially resolved developmental transitions. Informed, scalable and unified transport couplings therefore link atlas alignment to downstream biological inference within a single framework.