Performance of robust regression methods in real-time polymerase chain reaction calibration

Area of Science:

• Biostatistics
• Molecular Biology
• Analytical Chemistry

Background:

• Ordinary Least Squares (OLS) is standard for estimating nucleic acid concentrations via calibration.
• Outliers in quantitative real-time PCR (qPCR) data can compromise OLS accuracy.
• Robust regression methods are alternatives when data contain outliers.

Purpose of the Study:

• To compare the performance of OLS, Least Absolute Deviation (LAD), and biweight MM-estimator for qPCR calibration.
• To evaluate method performance in the presence and absence of outliers.
• To assess the trade-off between interval width and coverage probability.

Main Methods:

• Monte Carlo simulation was used to compare OLS, LAD, and MM-estimator.
• Outliers were introduced via replacement contamination in simulated qPCR calibration data.
• Performance was evaluated based on interval width and coverage probability.

Main Results:

• Without outliers, OLS and MM-estimator provided acceptable coverage and small widths; LAD had acceptable coverage but wider intervals.
• With outliers, OLS performance degraded significantly.
• MM-estimator maintained short interval widths but with reduced coverage, while LAD offered acceptable coverage at the cost of wider intervals.

Conclusions:

• Robust methods like MM-estimator and LAD are valuable for qPCR calibration with outlier-prone data.
• MM-estimator offers a balance between interval width and coverage in the presence of outliers.
• LAD provides robust coverage but with wider, less precise estimates.