Functionally Guided Graph Learning for Robust Cross-Patient Cell-Type Annotation in Single-Cell RNA Sequencing

Clinical Snapshot

Key Findings

• P Value: Not reported

• Effect Size: Average ACC of 84.28% and average F1-score of 84.08% across all 32 transfer tasks; ablation without biosemantic graph learning module yielded average ACC of 83.48% (absolute difference: 0.80 percentage points)

• Primary Outcome: Cell-type annotation accuracy (ACC) and F1-score across 32 directed cross-patient reference-to-query transfer tasks in three cancer scRNA-seq datasets (leukemia, breast invasive carcinoma, colorectal cancer)

• Nnt Or Sensitivity: Not applicable in the traditional clinical sense; classification performance metrics (ACC, F1-score) are the primary measures. Sensitivity and specificity per cell type are not reported. The 0.80 percentage point accuracy gain from the biosemantic module represents the key incremental effect size for the novel component.

• Confidence Interval: Not reported

Clinical Application

The method requires computational infrastructure capable of graph neural network training and KEGG pathway integration. Publicly available source code lowers the barrier to adoption for bioinformatics-capable laboratories. However, clinical implementation would require integration into existing scRNA-seq pipelines, validation on locally generated data, and assessment of runtime on clinical-scale datasets. The method is not currently validated for clinical diagnostic use. Australian clinical genomics laboratories (e.g., those operating under NATA accreditation for genomic testing) and research institutions participating in the Australian Genomics Health Alliance (AGHA) or cancer biobanking initiatives may find this method relevant for research-grade scRNA-seq analysis. There is no current TGA regulatory pathway specifically for scRNA-seq annotation software as a medical device, and PBS reimbursement for scRNA-seq-based diagnostics remains limited. RCPA and RACGP guidelines do not yet address scRNA-seq cell-type annotation methodology. Australian adoption would be confined to research and translational oncology settings. The method's reliance on KEGG pathway data — an internationally maintained resource — is compatible with Australian bioinformatics infrastructure. Researchers and clinical laboratories performing scRNA-seq-based cell-type annotation in cancer patients, particularly in haematological malignancies (leukemia), breast cancer, and colorectal cancer contexts where cross-patient annotation transfer is required

Abstract

Cross-patient cell-type annotation in single-cell RNA sequencing (scRNA-seq) remains challenging due to pronounced interpatient heterogeneity and distribution shifts across patient-specific cellular contexts. Conventional annotation approaches often rely on proximity-driven graph construction or expression similarity, which may introduce spurious cell-cell connections and lead to unstable knowledge transfer across patients. To address this limitation, we propose PathoGraph, a functionally guided graph learning framework for robust cross-patient cell-type annotation. The proposed method integrates KEGG-7-based biosemantic graph structure learning with cross-patient representation adaptation. Specifically, pathway-derived functional semantic profiles are incorporated to refine patient-specific cell graphs, encouraging biologically coherent neighborhoods and suppressing noise introduced by purely expression-based similarity. Based on the refined graphs, a cross-patient representation adaptation mechanism further aligns embeddings between labeled reference patients and unlabeled query patients to facilitate reliable annotation transfer. Experiments on three cross-patient scRNA-seq data sets, including leukemia, breast invasive carcinoma, and colorectal cancer data sets, demonstrate that PathoGraph achieves stable annotation performance across 32 directed reference-to-query transfer tasks. Across all tasks, PathoGraph obtained an average ACC of 84.28% and an F1-score of 84.08%, showing competitive and stable performance compared with representative marker-based, correlation-based, and model-based annotation methods. Ablation studies further show that removing the biosemantic graph learning module reduces the average accuracy to 83.48%, highlighting the importance of functional-guided graph refinement. In addition, post hoc functional relevance analyses in immune-cell and cancer-associated contexts suggest that the learned cell-cell graphs capture biologically relevant neighborhood structures beyond expression-driven proximity. The source code and processed data are publicly available at: https://github.com/LiYuechao1998/PathoGraph.

References

1. 1.Li, Y.-C., Wei, M.-M., Wang, X.-F., Wang, Z., Pan, J., Wang, L., Huang, Y.-A., Huang, Z.-A., & You, Z.-H. (2026). Functionally guided graph learning for robust cross-patient cell-type annotation in single-cell RNA sequencing. Journal of Chemical Information and Modeling. Advance online publication. https://doi.org/10.1021/acs.jcim.6c01857