scYOU: a multi-scale yoked optimization and unsupervised framework for analysis of cellular heterogeneity and complex-associated protein module recovery.

scYOU jointly learns cell and protein embeddings from proteomic data, using supercell-constrained alignment and GO-regularized protein priors to improve clustering and embedding separability.

• Why it matters: Current methods often treat cellular organization and protein structure separately, leaving no unified model that captures both heterogeneity and functional programs together.
• What they did: scYOU uses a dual-channel neural network with supercells built from high-order expression similarity, then applies a reconstruction-level alignment objective plus Gene Ontology-derived functional similarity priors.
• The result: On diverse benchmark datasets, scYOU shows strong, relatively consistent clustering and improved embedding separability, and its protein representations enable embedding-guided refinement of interaction networks and recovery of complex-associated protein modules.

AI summary of the abstract · may contain errors · not medical advice · read the paper.