Cross-level Cross-Scale Inference and Imputation of Single-cell Spatial Proteomics

Area of Science:

• Genomics
• Proteomics
• Computational Biology

Background:

• Single-cell and spatial omics technologies have advanced biological research but face challenges.
• Limitations include batch effects, lack of multi-modal measurements, limited protein coverage, and poor generalization.
• Insufficient spatial context at single-cell resolution hinders understanding molecular drivers.

Purpose of the Study:

• Introduce scProSpatial, a unified deep learning framework.
• Infer and impute high-fidelity single-cell spatial proteomics from scRNA-seq.
• Address limitations of current experimental omics methods.

Main Methods:

• Developed scProSpatial, a multi-modal, multi-scale deep learning framework.
• Framework infers and imputes spatial proteomics from single-cell RNA sequencing (scRNA-seq) data.
• Utilized comprehensive evaluations and a case study in metastatic breast cancer.

Main Results:

• scProSpatial accurately predicts spatial protein abundances without shared transcriptomics features.
• Expanded protein coverage by 50 times compared to existing methods.
• Demonstrated robust generalization to out-of-distribution scenarios.

Conclusions:

• scProSpatial effectively overcomes key challenges in single-cell spatial proteomics.
• The framework facilitates cross-level and cross-scale multi-omics integration.
• Enables deeper insights into complex biological systems, such as metastatic breast cancer.