DMS/NIGMS 1: Statistical modeling and estimation of cellular population dynamics

细胞培养分析是一种关键的实验方法，用于确定一组实验条件如何影响 细胞群体在体外的生长和动态。用于扰乱条件的方法包括改变 测量反应的干扰素或任何其他培养条件的量。为了量化这种关系 在所测试的培养条件和种群增长动态的变化之间，采用了统计模型 来估计和预测这些影响。目前的方法将细胞计数作为统计模型中的响应变量 并以IC50等指标总结结果。这些方法并不随时间、播种计数、 或其他条件，并可能导致重复性问题。不同的条件会导致不同的结果 相对细胞计数，即使生长动态相同，因此结果不容易推广到其他 场景。在这里，我们提出了一种严格的基于机制的响应估计方法，它使用 包含细胞分裂、死亡和状态间转换的种群增长的数学模型。利率 事件的响应是分层统计模型中的响应，该模型允许条件甚至细胞系的可变性。这个 结果是一个分析平台，它将细胞固有属性而不是细胞计数视为结果，因此它们是 对试验持续时间、播种密度和其他因素不变。我们提出这一新的方法论作为一个独立的 用于分析任何细胞培养实验数据的框架。我们将细胞生长建模为一个分枝过程 描述单个单元格或单元格类型如何通过定义以下各项来划分、消亡或经历单元格状态转换 事件作为由利率参数化的随机变量。我们附加了一个费率的统计模型作为以下函数 感兴趣的协变量。单元格计数形式的数据通过分支过程连接到费率，我们使用 用贝叶斯方法逼近似然估计模型的参数值。这种方法 创建了一个严格的框架，用于根据所获得的增长率执行细胞响应的估计 FORM计数作为输入数据，更一般地用于分支过程参数的估计。我们将建立 统计方法有以下几个目的：(1)我们将开发贝叶斯方法和统计框架 细胞出生率和死亡率的估计。(2)我们将创建一个分层模型框架来说明细胞系和 实验效果，以帮助创建可重现的结果。(3)我们将为更复杂的分支开发建模 可以解释经历各种状态转变的种群的动态的过程，包括循环， 差异化和大小。

Cell culture assays are a critical experimental method used to determine how a set of experimental conditions affect the growth and dynamics of a cell population in vitro. Methods used to perturb the conditions include varying the amount of perturbagen or any other culture condition to measure response. In order to quantify the relationship between the culture conditions tested and the change in population growth dynamics, a statistical model is employed to estimate and predict these effects. Current methods treat cell count as the response variable in statistical models and summarize the result in metrics like the IC50. These methods are not invariant to changes in time, seeding count, or other conditions, and can lead to reproducibility issues. Different conditions lead to different results in terms of relative cell count even if the growth dynamics are the same, so results are not easily generalizable to other scenarios. Here we propose a rigorous mechanism-based method for estimation of response that uses a mathematical model for population growth incorporating cell division, death, and transitions between states. The rate of events are the response in hierarchical statistical models that allows variability in conditions and even cell lines. The result is an analysis platform that treats cell-intrinsic properties rather than cell count as outcomes so that they are invariant to experimental duration, seeding density, and other factors. We propose this novel methodology as a standalone framework for analysis of any cell culture experimental data. We model cell growth as a branching process that describes how an individual cell or type of cells divide, die, or undergo cell state transitions by defining each of these events as random variables parameterized by the rates. We attach a statistical model for the rates as a function of covariates of interest. Data in the form of cell counts connects to rates through the branching process, and we use Bayesian methods to approximate the likelihood and estimate the parameter values of the model. This approach creates a rigorous framework for performing estimation of cellular response as a function of the growth rates obtained from counts as input data, and more generally for estimation of branching process parameters. We will establish the statistical methods in the following aims: (1.) We will develop Bayesian methods and a statistical framework for estimation of cell birth and death rates. (2.) We will create a hierarchical model framework to account for cell line and experimental effects to help create reproducible results. (3.) We will develop modeling for more complicated branching processes that can account for dynamics of a population undergoing a variety of state transitions including cycling, differentiation, and size.