GPTBioInsightor-leveraging large language models for transparent scRAN-seq cell type annotations.

Area of Science:

• Bioinformatics and Computational Biology
• Genomics and Molecular Biology

Background:

• Large language models (LLMs) are increasingly used in life sciences but often lack transparency in single-cell RNA sequencing (scRNA-seq) analysis.
• Opaque LLM tools hinder reproducibility, peer review, and adoption in bioinformatics.

Purpose of the Study:

• To develop an interpretable LLM-powered assistant for scRNA-seq analysis.
• To provide transparent reasoning, confidence scores, and evidence for AI-driven annotations.

Main Methods:

• Developed GPTBioInsightor, an LLM-based assistant for scRNA-seq data analysis.
• Enabled step-by-step narration of the decision-making process for cell type, state, and pathway activity annotation.

Main Results:

• GPTBioInsightor achieved parity with expert manual curation on benchmark datasets (PBMC3K, pancreatic cancer).
• The tool provides transparent reasoning, confidence scores, and literature-based evidence for its annotations.
• Closed the interpretability gap in AI-assisted bioinformatics.

Conclusions:

• GPTBioInsightor enhances the reliability and auditability of AI in scRNA-seq analysis.
• The tool accelerates scientific discovery by fostering trust and facilitating collaboration between wet-lab biologists and computational scientists.
• Ensures reproducible and transparent AI-driven insights in complex biological data analysis.