(W1130-06-40) Validating Digital Droplet PCR (ddPCR) Assay with Zero-Inflated Poisson Model

Purpose: Gene therapy, cell therapy, and vaccine developments from the bench to bed have led to increased use of ddPCR in bioanalytical laboratories, especially given that the absolute quantity feature of ddPCR assays. The Global CRO Council in Bioanalysis (GCC) has published a white paper on recommendations on ddPCR assay validation via following GLP and GMP analytics guidelines, e.g., for accuracy and precision (A&P), ±50% bias and ≤80% CV for samples with 50 copies/20 μl ddPCR reaction, ±35% bias and ≤40% CV for samples with ≥50 copies/20 μl ddPCR reaction, and lowest copy number at which the statistical model predicts the assay can detect 95% of the time and higher than the observed contamination level (mean copy number detected in blank samples or NTC + 3.3-fold standard deviation). Particularly, for ddPCR, the criteria of the lowest concentration of an analyte (LLOQ) in a sample that can be quantitatively determined with suitable accuracy and precision has yet to reach an agreement. However, available statistical models for digital droplet PCR (ddPCR) assays are based on the proportion of negative beads (droplets), while such zero counts are prone to be inflated because of insufficient droplet generating and PCR efficiency thereby leading to spurious zero values, i.e., zero-inflated. An improved model, rather than a single-point Poisson estimation, is an unmet need. We sought to develop a better model to improve the quality of ddPCR assay validation, involving how to improve the reliability of accuracy, precision, and sensitivity.

Methods: Here we use simulated data and empirical data sets to assess distinct models, including Poisson regression and zero-inflated models alongside. Furthermore, we follow the GCC guideline to assess the performance of the zero-inflated Poisson model.

Results: Zero-inflated Poisson regression model outperforms single-point Poisson in improving accuracy, precision, and sensitivity by means of flowing GCC guidelines. Of particular note, for samples with quantities ranging from 16 to 50 copies/20 µl per ddPCR reaction, the zero-inflated Poisson model enables the A&P to satisfy ±35% bias and ≤40% CV. For up to 12 repeats, the LLOQ can reach 16 copies/20 µl per ddPCR reaction with the criteria of A&p with ±25% bias and ≤25% CV, while the NTC satisfies less or equal than 2 copies/20 µl per ddPCR reaction. We implemented the Zero-Inflated Poisson Regression method into an R package ZIProde (Zero-Inflated Poisson Regression of Digital Droplet PCR), which is under validation.

Conclusion: We propose applying the zero-inflated Poisson model can substantially improve ddPCR method performance, where related parameters can be determined during validation with calibrators. Meanwhile, applying multiple repeats for LLOQ and NTC enables better accuracy and precision assessment.

References: Recommendations on qPCR/ddPCR assay validation by GCC.BIOANALYSISVOL. 14, NO. 12<