Serum and plasma are widely used in proteomic biomarker discovery, but differences between their proteomes have hindered the integration of data from the two specimen types. Here, we describe a computational approach for bridging between serum and plasma proteomic measurements derived from the aptamer-based SomaScan assay. We aimed to enable cross-specimen data utilization in the context of the PROphet model designed to predict immunotherapy outcomes based on 388 plasma proteomic biomarkers. Proteomic profiling of 7289 proteins was performed on 177 matched serum-plasma sample pairs from cancer patients across three distinct cohorts. Remarkably, 91.6% of the proteins showed correlation (p-value < 0.05) between serum and plasma protein levels, highlighting the feasibility of serum-plasma bridging. Linear scaling factors derived from matched serum-plasma sample pairs were consistent across the three cohorts, suggesting that the scaling factors are generalizable. Notably, the PROphet model maintained its predictive power when applied to scaled serum proteomic measurements. Specifically, clinical benefit predictions and survival stratification based on scaled serum proteomic measurements were similar to those based on plasma proteomic measurements. Our study demonstrates the feasibility of generalizing plasma-based predictors to serum samples through appropriate bridging strategies, paving the way for integrating serum and plasma datasets.
Keywords
