Optimizing biomarker discovery with focus on low coefficient of variation in large-scale proteomics

Background

Coefficients of variation (CV) describe innate technical variation in high throughput molecular measurement platforms and are a standard metric for characterizing and monitoring assay precision.

Median CVs range from ~4.5% to 18.0% for immunoassay technology,¹ up to >30% for mass spectrometry,² ~5% for the SomaScan® Assay, and ~10% for the Olink Explore Assay. Large CVs can cause technical variability to overwhelm biological signal.

Univariate analyses were conducted with multiple datasets and simulated CVs to assess the impact of increased CVs on biomarker discovery.

Methods

• Sample size calculations were made to identify how many additional samples are needed to account for platform variability (CVs of 5%, 10%, 15%, and 20%) assuming a power of 80% and a significance level of 5%.
• SomaScan® assay v4.1 measured over 7,000 analytes in EDTA plasma for datasets with three biological outcomes : incident preeclampsia, maximum COVID-19 severity, and incident obstructive coronary artery disease (oCAD) (Table 1).
• CV increases were simulated by maintaining constant mean analyte values while adding normally distributed noise based on CVs from
  technical replicates. The mean results from 200 simulations were reported for each analyte.
• The simulation methods were verified by comparing CVs of technical replicates before and after adding noise (Table 2).
• The 2x and 3x increased CVs are in the range of the CVs for the midplex affinity-based 1K and 3K assays (band Table 2).

Conclusions

• Technical variability (higher CVs) decreases detectability of between-group differences, particularly when effect sizes are small.
• With higher CVs, either fewer biomarkers will be identified or more samples will be required.
• By utilizing a platform with low CVs, sample efficiency and opportunity for biomarker discovery are maximized.