scmFormer Integrates Large-Scale Single-Cell Proteomics and Transcriptomics Data by Multi-Task Transformer.

Area of Science:

• Computational Biology
• Genomics
• Proteomics

Background:

• Transformer models have advanced single-cell RNA sequencing (scRNA-seq) analysis.
• Integrating large-scale and complex single-cell multi-omics data, especially proteomics, remains a challenge.

Purpose of the Study:

• To introduce scmFormer, a novel single-cell multi-modal/multi-task transformer.
• To address the gap in integrating single-cell proteomics with other omics data.

Main Methods:

• Systematic benchmarking of scmFormer on large-scale multimodal and multi-batch omics data.
• Evaluation of scmFormer's performance in cell-type label transfer from transcriptomics to proteomics.
• Assessment of scmFormer's capability in generating unmeasured modalities and handling spatial multi-omic data.

Main Results:

• scmFormer excels at integrating large-scale, heterogeneous, multi-batch single-cell multi-omics data.
• Achieved a 54.5% higher average F1 score in cell-type label transfer compared to the next best method.
• Successfully integrated over 1.48 million cells from COVID-19 datasets on a personal computer.
• Demonstrated superior performance in generating unmeasured data modalities and suitability for spatial multi-omics.

Conclusions:

• scmFormer is a powerful and comprehensive tool for single-cell multi-omics data integration.
• The model effectively preserves shared and distinct biological information across batches and modalities.
• scmFormer demonstrates scalability and efficiency for analyzing massive single-cell datasets.