Journal: Bioinformatics Advances
Article Title: Cell type annotation using large language models (LLMs) and CytoAnalyst
doi: 10.1093/bioadv/vbag001
Figure Lengend Snippet: LLM-powered cell annotation for single-cell RNA sequencing data (scRNA-Seq). (A) Data upload and quality control. Data upload imports single-cell data (in 10X Genomics Cell Ranger or AnnData format) and metadata, while quality control filters out low-quality cells and genes. (B) Marker discovery pipeline, including: (1) embedding analysis, (2) visualization, (3) clustering, and (4) marker discovery through interactive differential analysis. (C) LLM-powered cell type inference and interactive annotation. The inference workflow uses a structured prompt template that guides the LLM to predict potential cell types using the provided gene sets, tissue information, and cell ontology, ensuring biologically meaningful predictions. The interactive annotation interface allows users to combine automatic annotation and domain expertise with advanced cell filtering capabilities. (D1–D5) Analysis results of the case study using bone marrow organoids. The left panels show the inferred lineage hierarchies and predicted cell types, while the right panels show the expression patterns of marker genes for the five cell groups identified by the platform.
Article Snippet: Accepted input includes 10X Genomics Cell Ranger output (.tar.gz or .h5) and AnnData (.h5ad) objects, along with optional metadata containing sample information and experimental conditions for cell type identification.
Techniques: Single Cell, RNA Sequencing, Control, Marker, Expressing