Scalable Multi-omics Diagonal Integration from Self-Mapping

SCALEMAP enables scalable diagonal integration of single-cell multi-omics by self-mapping cross-modality predictions to each modality’s own observed inputs, integrating 1.5 million profiles in under 5 minutes.

• Why it matters: Diagonal integration is hard because modalities are measured in different cells and lack direct correspondences, making weak-linkage anchor search and refinement computationally burdensome at atlas scale.
• What they did: The method uses a self-mapping mechanism that aligns modalities by mapping each decoded cross-modality omics profile prediction back to its own observed inputs, implemented as SCALEMAP across transcriptomic, epigenomic, and proteomic benchmarks.
• The result: Across benchmarks, SCALEMAP matches or outperforms leading alternatives while using substantially less runtime and memory, achieving integration with under 20 GB peak memory.

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