Benchmarking digital PCR partition classification methods with empirical and simulated duplex data.

Area of Science:

• Biotechnology
• Molecular Biology
• Bioinformatics

Background:

• Digital PCR (dPCR) offers high accuracy for nucleic acid quantification, with significant clinical potential in areas like tumor liquid biopsy and biomarker validation.
• Precise classification of partitions based on fluorescence is critical for unbiased concentration estimation in dPCR.
• Existing clustering methods may not adequately address the complexities of dPCR data, potentially leading to biased results.

Purpose of the Study:

• To comprehensively evaluate various clustering methods for dPCR data analysis.
• To develop guidelines for selecting appropriate clustering methods for dPCR applications.
• To introduce a novel, realistic dPCR data simulation method and a corresponding R Shiny app.

Main Methods:

• Evaluation of general-purpose, dPCR-specific, and flow cytometry clustering methods using simulated and real dPCR data.
• Performance assessment of clustering methods across diverse simulated scenarios.
• Development of a new dPCR data simulation method based on a mixture distribution (Poisson point process and skew-t distribution).

Main Results:

• Identification of the limitations of various clustering methods for dPCR data.
• Formulation of evidence-based guidelines for choosing optimal clustering approaches.
• Successful development of a novel simulation method that generates realistic dPCR data with common irregularities like 'rain'.

Conclusions:

• The choice of clustering method significantly impacts dPCR quantification accuracy.
• The developed simulation tool and guidelines provide valuable resources for dPCR data analysis and method development.
• The R Shiny app facilitates the generation of labeled dPCR datasets for training and testing new algorithms.