Health Sciences and Technology, Harvard Medical School, Boston, MA 02115, USA.
Bioinformatics. 2011 Oct 1;27(19):2746-53. doi: 10.1093/bioinformatics/btr468. Epub 2011 Aug 16.
Monoclonal antibodies (mAbs) are among the most powerful and important tools in biology and medicine. MAb development is of great significance to many research and clinical applications. Therefore, objective mAb classification is essential for categorizing and comparing mAb panels based on their reactivity patterns in different cellular species. However, typical flow cytometric mAb profiles present unique modeling challenges with their non-Gaussian features and intersample variations. It makes accurate mAb classification difficult to do with the currently used kernel-based or hierarchical clustering techniques.
To address these challenges, in the present study we developed a formal two-step framework called mAbprofiler for systematic, parametric characterization of mAb profiles. Further, we measured the reactivity of hundreds of new antibodies in diverse tissues using flow cytometry, which we successfully classified using mAbprofiler. First, mAbprofiler fits a mAb's flow cytometric histogram with a finite mixture model of skew t distributions that is robust against non-Gaussian features, and constructs a precise, smooth and mathematically rigorous profile. Then it performs novel curve clustering of the fitted mAb profiles using a skew t mixture of non-linear regression model that can handle intersample variation. Thus, mAbprofiler provides a new framework for identifying robust mAb classes, all well defined by distinct parametric templates, which can be used for classifying new mAb samples. We validated our classification results both computationally and empirically using mAb profiles of known classification.
A demonstration code in R is available at the journal website. The R code implementing the full framework is available from the author website - http://amath.nchu.edu.tw/www/teacher/tilin/software
saumyadipta_pyne@dfci.harvard.edu
Supplementary data are available at Bioinformatics online.
单克隆抗体 (mAb) 是生物学和医学中最强大和最重要的工具之一。 mAb 的开发对于许多研究和临床应用都具有重要意义。因此，客观的 mAb 分类对于根据其在不同细胞物种中的反应模式对 mAb 面板进行分类和比较至关重要。然而，典型的流式细胞术 mAb 谱具有独特的建模挑战，具有非高斯特征和样本间变化。这使得使用当前基于核的或层次聚类技术难以进行准确的 mAb 分类。
为了解决这些挑战，在本研究中，我们开发了一种称为 mAbprofiler 的正式两步框架，用于对 mAb 谱进行系统、参数化表征。此外，我们使用流式细胞术测量了数百种新抗体在不同组织中的反应性，并成功使用 mAbprofiler 对其进行了分类。首先，mAbprofiler 使用斜 t 分布的有限混合模型拟合 mAb 的流式细胞术直方图，该模型对非高斯特征具有鲁棒性，并构建了精确、平滑且数学上严格的谱。然后，它使用非线性回归模型的斜 t 混合对拟合的 mAb 谱进行新的曲线聚类，该模型可以处理样本间变化。因此，mAbprofiler 提供了一种新的框架，用于识别稳健的 mAb 类，所有这些类都由独特的参数模板定义，可以用于对新的 mAb 样本进行分类。我们使用已知分类的 mAb 谱在计算和经验上验证了我们的分类结果。
在期刊网站上提供了 R 中的演示代码。实现完整框架的 R 代码可从作者网站获得 - http://amath.nchu.edu.tw/www/teacher/tilin/software
saumyadipta_pyne@dfci.harvard.edu
补充数据可在 Bioinformatics 在线获得。
