Southeast Ecological Science Center, U.S. Geological Survey , 7920 NW 71st Street, Gainesville, Florida 32653, United States.
Anal Chem. 2015 Nov 3;87(21):10886-93. doi: 10.1021/acs.analchem.5b02429. Epub 2015 Oct 13.
Statistical methods for the analysis and design of experiments using digital PCR (dPCR) have received only limited attention and have been misused in many instances. To address this issue and to provide a more general approach to the analysis of dPCR data, we describe a class of statistical models for the analysis and design of experiments that require quantification of nucleic acids. These models are mathematically equivalent to generalized linear models of binomial responses that include a complementary, log-log link function and an offset that is dependent on the dPCR partition volume. These models are both versatile and easy to fit using conventional statistical software. Covariates can be used to specify different sources of variation in nucleic acid concentration, and a model's parameters can be used to quantify the effects of these covariates. For purposes of illustration, we analyzed dPCR data from different types of experiments, including serial dilution, evaluation of copy number variation, and quantification of gene expression. We also showed how these models can be used to help design dPCR experiments, as in selection of sample sizes needed to achieve desired levels of precision in estimates of nucleic acid concentration or to detect differences in concentration among treatments with prescribed levels of statistical power.
使用数字 PCR（dPCR）进行实验分析和设计的统计方法仅受到有限关注，并且在许多情况下被误用。为了解决这个问题，并为 dPCR 数据的分析提供更通用的方法，我们描述了一类用于需要定量核酸的实验分析和设计的统计模型。这些模型在数学上等同于包含互补的对数-对数链接函数和与 dPCR 分区体积相关的偏移量的二项式响应的广义线性模型。这些模型既通用又易于使用常规统计软件进行拟合。协变量可用于指定核酸浓度变化的不同来源，并且模型的参数可用于量化这些协变量的影响。为了说明问题，我们分析了来自不同类型实验的 dPCR 数据，包括系列稀释、拷贝数变异评估和基因表达定量。我们还展示了如何使用这些模型帮助设计 dPCR 实验，例如选择样本量以达到所需的核酸浓度估计精度，或在具有预定统计功效水平的处理之间检测浓度差异。
