The question
Vector copy number (VCN) is measured by qPCR at one lab and by ddPCR at another, and the qPCR results run consistently higher. Is one method wrong, and does it matter against a release limit of 5 copies per cell?
The analysis
| qPCR calibration | Slope (95% CI) | Intercept (95% CI) | Mean % difference |
|---|---|---|---|
| Supercoiled standard | 1.32 (1.22 to 1.41) | -0.01 (-0.33 to 0.27) | 26.9% |
| Linearized standard | 0.95 (0.90 to 1.01) | 0.15 (-0.01 to 0.34) | 1.0% |
With the supercoiled plasmid standard, the Passing-Bablok slope confidence interval sits well above 1 while the intercept interval includes 0. That is a proportional bias: qPCR reads a fixed percentage high across the whole range, not a fixed amount high. A proportional bias points at calibration, and the usual suspect is the standard itself. Supercoiled plasmid amplifies less efficiently than linear genomic targets, which shifts the standard curve and inflates every sample reading by the same factor.
What to do
- Linearize the plasmid standard (restriction digest) or anchor it to a ddPCR-quantified reference, then re-run the comparison; the corrected set above agrees with ddPCR.
- Rule out residual unintegrated vector as a second contributor by testing samples at more than one time after transduction.
- Decide which method is the reference for release before comparing them, and write the bridging criteria down first.
How it was done
Synthetic paired VCN results for 24 lots. Passing-Bablok regression with 95% confidence intervals; Bland-Altman on the percent scale with 95% limits of agreement. You can run the same analysis on your own data in the method comparison tool. Download the data (CSV).