1. 연사 :  이미성 교수 (단국대 통계데이터사이언스학과)

2. 주제 : Functional Protein Biomarkers Based on Distributions of Single-Cell Expression Levels: From Linear to Nonlinear Quantile Index Predictors

3. 날짜 : 2025년 12월 22일(월) 16:00

4. 장소 : 사이언스홀 (자1-100호)

5. 발표 초록 :  

Background: Protein biomarkers of cancer progression and response to therapy are increasingly important for personalized medicine. Current histocytometry methods enable single-cell quantification of biomolecules in tumor tissue sections through multiplex immunohistochemistr. Advanced quantitative pathology platforms provide distributions of cellular signal intensity (CSI) levels across entire cell populations. However, this rich cell-by-cell biomarker information is typically reduced to a single mean or converted into a simple proportion of biomarker-positive cells, failing to exploit intra-tumor heterogeneity that may provide important prognostic information.

Methods: We developed a comprehensive framework using distributions of functional single-cell protein expression levels as cancer biomarkers. The quantile index (QI) biomarker is defined as a weighted average of CSI distribution quantiles in individual tumors, where the weight for each quantile is determined by fitting a functional regression model for a clinical outcome. We extended this to nonlinear QI (nlQI) biomarkers allowing the association between the CSI quantile function and outcome to vary nonlinearly. An algorithm was developed for selecting optimal cutoffs for dichotomizing cell signal intensity distribution quantiles as predictors of continuous, categorical, or survival outcomes. For many functional proteins, single-cell expressions vary independently of spatial localization, and incorporation of spatial information may not affect prognostic value.

Results: Linear and nonlinear QI biomarkers based on single-cell expressions of ER, Ki67, TS, CyclinD3, PCNA, PD-L2, and PR were derived and evaluated as predictors of progression-free survival (PFS) or high mitotic index in a large breast cancer dataset. The QI biomarkers demonstrated improved prognostic value compared with standard mean signal intensity predictors. Performance was validated using an independent external cohort. For proteins significantly associated with PFS, optimal quantile biomarkers yielded either larger or similar effect sizes as compared to mean signal intensity biomarkers. Simulation studies demonstrated that nlQI biomarkers yield higher predictive power than linear QI biomarkers when between-tissue variability in CSI distributions is substantial.

Conclusions: The proposed approach can be applied to any cell-level expressions of proteins or nucleic acids from immunohistochemistry or other single-cell technologies. R packages Qindex and hyper.gam implementing these methods are freely available on CRAN, featuring user-friendly interfaces and visual tools for exploring integrand surfaces. These tools address the need for biomarkers accounting for heterogeneous protein expression levels in tissues.

Keywords: Quantile index, Single-cell imaging, Multiplex immunofluorescence, Distribution quantiles, Protein biomarker, Functional regression

This work was supported by the National Institutes of Health, U.S. Department of Health and Human Services grants R01CA222847 (I.C., T.Z., and H.R.) and R01CA253977 (H.R. and I.C.). Generation of the underlying data was supported by a Komen Promise grant KG091116 awarded to a team of investigators led by H.R. and the late Edith P. Mitchell.