A joint analysis of single cell transcriptomics and proteomics using transformer.

Area of Science:

• Single-cell biology
• Computational biology
• Molecular biology

Background:

• CITE-seq simultaneously measures RNA and protein expression at the single-cell level, crucial for understanding cellular heterogeneity.
• High experimental costs of CITE-seq limit its broad application.
• Integrated analysis of RNA and protein expression is vital for deep biological insights.

Purpose of the Study:

• To develop a computational method, scTEL, for predicting protein expression from single-cell RNA sequencing (scRNA-seq) data.
• To reduce the experimental costs associated with protein expression profiling.
• To create a unified framework for integrating multiple CITE-seq datasets with varying protein panels.

Main Methods:

• scTEL utilizes a deep learning framework based on Transformer encoder layers.
• The model establishes a mapping from measured RNA expression to unobserved protein expression within the same cells.
• A unified framework is proposed to integrate multiple CITE-seq datasets, handling partial overlap in protein panels.

Main Results:

• scTEL accurately predicts protein expression from cost-effective scRNA-seq data.
• The computational approach significantly lowers the cost of protein expression analysis.
• The model demonstrates superior performance compared to existing methods in empirical validation on public CITE-seq datasets.
• scTEL effectively integrates multiple CITE-seq datasets with heterogeneous protein panels.

Conclusions:

• scTEL offers a cost-effective computational solution for inferring protein expression from scRNA-seq data.
• The framework facilitates the integration of diverse CITE-seq datasets, enhancing multi-omic single-cell analysis.
• This approach democratizes high-resolution cellular profiling by reducing experimental barriers.