[FPCAM: a weighted dictionary-based single-cell annotation model for pulmonary fibrosis]

The breakthrough development of single-cell RNA sequencing (scRNA-seq) has provided unprecedented resolution for analyzing cellular heterogeneity. However, challenges remain in the efficiency and accuracy of cell type annotation. Existing automated annotation tools are limited by their dependence on reference datasets, polymorphism of marker genes, and subjective bias introduced by manual intervention, which make it difficult to accurately annotate complex cell subpopulations and cross-platform datasets. To address these challenges, we developed FPCAM, a fully automated cell annotation tool based on the R Shiny platform. This model leverages the Seurat framework, utilizing FindAllMarkers to identify feature genes and integrating similarity matrix calculations, a manually curated pulmonary fibrosis-related cell-gene association dictionary, and an optimized evaluation metric algorithm to achieve efficient single-cell annotation. To assess the model performance, we compared FPCAM with the most advanced annotation models, including SCSA, SingleR, and SciBet. The results showed that FPCAM achieved the accuracy of 85.7%, outperforming SCSA (82.1%) and SciBet (78.6%). FPCAM showed the Cohen's Kappa coefficient reaching 0.81, also outperforming SCSA (0.76) and SciBet (0.73). Additionally, the ranges of SingleR and SciBet were 0.357 and 0.322, respectively, indicating their strong dependence on predefined cell annotation files, while FPCAM demonstrated superior accuracy and stability in annotation tasks. Overall, FPCAM integrates a multi-source marker gene database with a dynamic update strategy and employs an innovative weighted annotation algorithm to achieve efficient, flexible, and precise cell-type identification. It serves as a powerful tool for single-cell transcriptomic studies of pulmonary fibrosis and other diseases.