Reducing Bias in Digital PCR Quantification Experiments: The Importance of Appropriately Modeling Volume Variability

Multiple simulation studies have shown that volume variability of partition sizes in digital PCR (dPCR) causes bias in the resulting concentration estimates and their associated standard errors. These biases are especially apparent when the volume variability is large, and the targeted nucleic acid concentration is high. Currently, only a single method for the elimination or reduction of these biases is available, and it assumes a fixed class of models for the volume variability. We show that the form in which volumetric variability occurs in empirical data is variable and cannot be modeled by a single distribution. We propose a new volume-modeling method, NPVolMod, which takes volume variability of an arbitrary form into account and is applicable to both absolute and relative quantification. The method is nonparametric in the sense that no distributional assumption is needed. Moreover, the volumes of each of the individual partitions are not needed. We empirically demonstrate by simulation that NPVolMod nearly eliminates the biases caused by volumetric variability and that it often outperforms the existing method. The possibility of the proper modeling of volume variability may have implications for platform design and may increase the performance of existing dPCR platforms in terms of, for example, their trueness and linear dynamic range.