Enhancing cell type annotation for cancer transcriptomics using retrieval-augmented generation.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for cancer research, revealing cellular heterogeneity.
• Accurate cell type annotation is a key challenge, impacting downstream analysis reliability.

Purpose of the Study:

• To develop a robust retrieval-augmented generation framework for automated cell type annotation.
• To improve the stability, biological grounding, and reproducibility of scRNA-seq data analysis.

Main Methods:

• Integrated hybrid lexical-semantic information retrieval with ontology-aware evaluation.
• Utilized keyword search, biomedical embeddings, query expansion, and Cell Ontology-grounded scoring.
• Evaluated across nine diverse human tissue scRNA-seq datasets.

Main Results:

• The framework demonstrated consistently higher annotation accuracy compared to baseline methods.
• Showed reduced performance variability and improved resolution of rare and ambiguous cell populations.
• Enhanced biological grounding and reproducibility in cell type annotation.

Conclusions:

• The developed framework significantly improves automated cell type annotation for scRNA-seq data.
• Facilitates more accurate analysis of genetic alterations in cancer biology.
• Provides a scalable foundation for translational and diagnostic applications in cancer genetics.