Biomedical Signal Processing and Control Обработка сигналов

Confidence-gated forecasting of qPCR amplification curves for adaptive reporting

Zhenwei Wu, Daojun Yu, Xueyan Dong, Jian Wu
Elsevier · · DOI 10.2139/ssrn.7471135

Аннотация

Quantitative polymerase chain reaction (qPCR) analysis generally relies on sufficiently developed amplification curves, limiting the use of curve information before acquisition is complete. We developed a sequential, confidence-gated framework to forecast the unobserved segment of a qPCR amplification curve from partial fluorescence measurements. Results are reported at the first observation cycle satisfying validation-defined criteria for both classification probability and curve confidence. The framework was developed using 6,305 clinical hepatitis B virus qPCR records divided chronologically into training, validation, and later-period test cohorts. In the later-period test cohort (n = 1,264), positive and negative reactions were reported after means of 16.56 and 22.35 observation cycles, respectively. At these reporting cycles, classification accuracy was 99.68%, and the reconstructed curves had a median complete-curve R² of 0.9841. At a comparable reporting cycle, confidence-gated reporting improved classification accuracy and reduced future-segment error relative to fixed-cycle reporting. Future-segment forecasts also outperformed conventional logistic and Gompertz extrapolation at all evaluated observation cycles. Quantification-cycle (Cq) estimates derived from reconstructed curves showed mean absolute errors of 0.29 cycles with LinRegPCR and 0.49 cycles with Cy0 relative to estimates from measured curves. When applied without retraining, the model achieved classification accuracies of 99.71% and 97.59% in the two additional datasets. These results show that partial qPCR curves can support curve reconstruction and reaction-specific reporting, with potential for broader quantitative and kinetic analyses.

Читать у издателя →