The Digital PCR consortium at Ghent University offers services in experimental design, assay design, digital PCR, and statistical data analysis. Ward De Spiegelaere, Associate Professor, Ghent University Digital PCR Consortium discussed digital PCR classification and the various methods that can be used for this.
When classifying partitions, De Spiegelaere advised researchers to look at existing methods before designing their own tools. There are three main assumptions in digital PCR: random partitioning, equal partition size, and unambiguous classification (positive/negative) of partitions. In theory, digital PCR is a straightforward process but in practice fluorescence intensity variations and the presence of “rain” partitions complicate the classification process.
De Spiegelaere advocated the use of automated methods to prevent user bias and conduct high-throughput analysis, and high multiplexing in digital PCR. He also mentioned that linear and non-linear methods can be applied to dPCR.
Regarding high multiplexing in dPCR, multiple targets can be analysed using fluorescence colours and combining them in different ways and proportions. Although this technique improves digital PCR capabilities, the issue of cluster separation prevails. De Spiegelaere mentioned that his consortium is collaborating with PXLENCE, a spin-off company from Ghent University to develop rainbow probes. Rainbow probes use generic probes linked to primers rather than target-specific sequences to simplify multiplexing.
Moving to the hurdles, well-to-well variation, baseline shifts, and non-orthogonal cluster positions contribute to partition classification difficulty. Most classification methods only perform well in data sets but struggle with variation. K-means clustering, DBSCAN, and FlowPeaks were recognised as the best-performing techniques.
The team has developed simulation models to generate synthetic dPCR datasets with controlled variables, this enables a systematic evaluation of classification methods. De Spiegelaere concluded that since no universal digital PCR classification method, further work in AI-driven and statistical modelling is expected to enhance dPCR data analysis.
