A natural language processing system for the efficient extraction of cell markers.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables detailed cellular exploration but requires accurate cell type annotation.
• Current cell annotation relies on manual curation and existing marker databases, which can be incomplete.

Purpose of the Study:

• To develop and validate MarkerGeneBERT, a natural language processing (NLP) system for extracting cell type and marker gene information from scRNA-seq literature.
• To create a comprehensive database of cell markers and types from scientific publications.

Main Methods:

• Developed MarkerGeneBERT, an NLP system, to parse full-text articles from 3702 scRNA-seq studies.
• Extracted species, tissue, cell type, and cell marker gene information.
• Validated extracted data against manually curated databases and applied it to scRNA-seq data annotation.

Main Results:

• Collected 7901 human cell markers (1606 types) and 8223 mouse cell markers (1674 types).
• Achieved 76% completeness and 75% accuracy compared to existing databases.
• Identified 89 novel cell types and 183 new marker genes.
• Successfully annotated brain tissue scRNA-seq data using compiled markers, consistent with original studies.

Conclusions:

• NLP-based methods, like MarkerGeneBERT, significantly expedite and improve scRNA-seq data annotation and interpretation.
• MarkerGeneBERT offers a powerful, scalable approach to augment biological databases and advance single-cell research.
• The system demonstrates transformative potential for understanding cellular heterogeneity.