PO.BCS01.17 · 生物信息与计算
用于空间蛋白质成像数据分析的三种细胞群共定位测量方法
Measure of three cell population co-localization for spatial protein imaging data analysis
作者与单位 Authors & Affiliations
摘要 Abstract
中文摘要
背景：空间蛋白质成像技术能够对肿瘤微环境（TME）进行细致研究，以刻画细胞丰度和空间结构。成对共定位指标（如Ripley's K）会遗漏高阶模式，而对三种细胞群枚举所有三角形在全切片图像上的计算是不可行的。使用三角形最长边的指标难以区分紧凑三元组与拉长或分散的排布。因此，我们提出一种使用三角形面积的新型三变量共定位测量方法。
方法：我们开发了一种Horvitz-Thompson蒙特卡洛估计量，并在模拟数据上将未加权和等周加权的面积变体与最长边类比方法进行基准比较。这些指标应用于一项原发性高级别浆液性癌（n=101）的研究，以量化T细胞（CD3+）、细胞毒性T细胞（CD3+CD8+）、B细胞（CD19+）和巨噬细胞（CD68+）之间的共定位。Cox比例风险（PH）模型在15和25 μm半径下评估了标准化聚集指标与总生存期之间的关联——校正了诊断年龄、分期和肿瘤减灭情况。
结果：模拟显示，基于面积的测量方法给出无偏、一致的共定位估计，且在小半径下方差更低。蒙特卡洛抽样相比完全枚举实现了>10^5倍的运行时间缩减。在该卵巢癌研究中，淋巴细胞与巨噬细胞较低的三细胞共定位水平对应于总生存期的小幅改善，尽管不显著（表1）。基于面积的估计量还观察到更窄的置信区间，加权和未加权面积变体表现相似。
讨论：我们的框架能够对TME中的高阶空间组织进行可扩展的量化。尽管在该卵巢癌研究中各估计量的关联相当，但基于面积的测量方法更高的稳定性支持其在未来空间研究中作为复杂细胞结构的替代指标。表1：所选半径下研究细胞三元组的Cox模型风险比及95%置信区间。 细胞三元组 半径 最长边 未加权面积 加权面积 T细胞
B细胞
巨噬细胞 15 0.83 (0.61, 1.14) 0.93 (0.66, 1.31) 0.89 (0.65, 1.21) 25 0.90 (0.65, 1.23) 0.91 (0.67, 1.23) 0.93 (0.68, 1.27) 细胞毒性T细胞
B细胞
巨噬细胞 15 0.95 (0.73, 1.23) 0.90 (0.68, 1.17) 0.93 (0.73, 1.20) 25 0.85 (0.65, 1.12) 0.86 (0.66, 1.11) 0.84 (0.64, 1.09)
查看英文原文 English abstract
Background: Spatial protein imaging technologies enable detailed study of the tumor microenvironment (TME) to characterize cell abundance and spatial architecture. Pairwise colocalization metrics (e.g., Ripley's K) miss higher-order patterns, while enumerating all triangles for three cell populations is computationally infeasible on whole-slide images. Metrics using a triangle's longest edge poorly distinguish compact triples from elongated or dispersed arrangements. Thus, we propose a new trivariate colocalization measure using triangle area.
Methods: We developed a Horvitz-Thompson Monte Carlo estimator and benchmarked the unweighted and isoperimetrically weighted area variants against a longest-edge analogue on simulated data. These metrics were applied to a study of primary high-grade serous carcinoma (n=101) to quantify colocalization among T cells (CD3+), cytotoxic T cells (CD3+CD8+), B cells (CD19+), and macrophages (CD68+). Cox PH models evaluated associations between normalized clustering metrics and overall survival at radii of 15 and 25 μm - adjusted for diagnosis age, stage, and debulking.
Results: Simulations showed unbiased, consistent colocalization estimates with lower small-radius variance for area-based measures. Monte Carlo sampling achieved a >10 5 -fold runtime reduction versus full enumeration. In the ovarian study, lower three-cell colocalization levels of lymphocytes with macrophages corresponded to small improvements in overall survival, though not significant (Table 1). Narrower confidence intervals were also observed for area-based estimators, with weighted and unweighted area variants performing similarly.
Discussion: Our framework enables scalable quantification of higher-order spatial organization in the TME. Despite comparable associations across estimators in the ovarian study, the greater stability of area-based measures supports their use as surrogates of complex cellular architecture in future spatial studies. Table 1: Cox model hazard ratios and 95% confidence intervals for study cell triples at chosen radii Cell Triple Radius Longest-Edge Unweighted Area Weighted Area T Cell
B Cell
Macrophage 15 0.83 (0.61, 1.14) 0.93 (0.66, 1.31) 0.89 (0.65, 1.21) 25 0.90 (0.65, 1.23) 0.91 (0.67, 1.23) 0.93 (0.68, 1.27) Cytotoxic T Cell
B Cell
Macrophage 15 0.95 (0.73, 1.23) 0.90 (0.68, 1.17) 0.93 (0.73, 1.20) 25 0.85 (0.65, 1.12) 0.86 (0.66, 1.11) 0.84 (0.64, 1.09)
利益披露 Disclosure
K. Sabitov, None..
A. Soupir, None..
L. C. Peres, None..
B. L. Fridley, None.