Abstract
Digital polymerase chain reaction (dPCR) is a state-of-the-art targeted quantification method of nucleic acids. The technology is based on massive partitioning of a reaction mixture into individual PCR reactions. The resulting partition-level end-point fluorescence intensities are used to classify partitions as positive or negative, i.e. containing or not containing the target nucleic acid(s). Many automatic dPCR partition classification methods have been proposed, but they are limited to the analysis of single- or dual-color dPCR data. While general-purpose or flow cytometry clustering methods can be directly applied to multicolor dPCR data, these methods do not exploit the approximate prior knowledge on cluster center locations available in dPCR data. We present Polytect, a method that relies on crude cluster results from flowPeaks, previously shown to offer good partition classification performance, and subsequently refines flowPeaks' results by automatic cluster merging and cluster labeling, exploiting the prior knowledge on cluster center locations. Comparative analyses with established methods such as flowPeaks, dpcp, and ddPCRclust reveal that Polytect often surpasses established methods, both on empirical and simulated data. Polytect manages to merge excess clusters, while also successfully identifying empty clusters when fewer than the maximally observable number of clusters are present. On par with recent developments in instruments, Polytect extends beyond two-color data. The method is available as an R package and R Shiny app (https://digpcr.shinyapps.io/Polytect/).