A pre-trained large generative model for translating single-cell transcriptomes to proteomes.

Area of Science:

• Single-cell biology
• Computational biology
• Proteomics

Background:

• Single-cell protein abundance measurement is crucial for understanding cellular processes and disease.
• Current single-cell proteomic technologies have limitations including low coverage, throughput, sensitivity, and high costs.

Purpose of the Study:

• To develop a pre-trained, large generative model, scTranslator, to infer missing single-cell proteomes from transcriptomic data.
• To address the challenges faced by existing single-cell proteomic technologies.

Main Methods:

• Utilized natural language processing and genetic central dogma principles to design scTranslator.
• Employed a pre-trained, large generative model approach for inferring proteomic data from transcriptomic data.
• Validated scTranslator across diverse datasets, profiling techniques, cell types, tissues, and disease contexts.

Main Results:

• scTranslator accurately generates multi-omics data by inferring single-cell proteomes from transcriptomes.
• Demonstrated the model's accuracy, stability, and flexibility across various experimental techniques (CITE-seq, spatial CITE-seq, REAP-seq, NEAT-seq) and biological contexts.
• Showcased scTranslator's effectiveness in downstream applications like gene/protein interaction inference, perturbation prediction, cell clustering, batch correction, and pan-cancer cell origin recognition.

Conclusions:

• scTranslator offers a powerful solution for enhancing single-cell multi-omics analysis by inferring proteomic data.
• The model overcomes key limitations of current proteomic technologies, enabling high-resolution biological insights.
• scTranslator demonstrates broad applicability and superiority in various downstream analyses across diverse biological and disease settings.