Multiplexed Aptamer-Based Proteomic Data Normalization (Adaptive Normalization by Maximum Likelihood) Significantly...

Blood-based proteomics are employed in cohorts of patients with critical illness to provide biologic insights and improve diagnosis and risk stratification. Somalogic developed a multiplex aptamer-based platform (Somascan) with thousands of proteins assayed in a single plasma sample. Results are reported in 2 forms: (a) adaptive normalization by maximum likelihood (ANML), which normalizes measurements to a healthy US population and (b) a non-ANML-normalized form. It is unknown how ANML-normalization influences measurements and associations with outcomes.

We correlated ANML and non-ANML protein quantification with 9 immunoassay biomarkers measured in the same patients' blood samples from the COVID-19 Host Response and Clinical Outcomes (CHROME) cohort. We compared the association of log2 relative fluorescent units (RFUs) with hospital mortality with and without ANML-normalization using adjusted (age, sex, body mass index [BMI]) logistic regression. Results report the odds ratio per doubling of each protein's RFU and declared significance at a false discovery rate (FDR) <0.05.

Among 197 patients the average age was 54 years (±16), 51% had COVID-19, 65% were male and hospital mortality was 26%. All biomarkers had equal or higher Pearson correlation between non-ANML-normalization and immunoassay data compared to ANML-normalization. Thousands of proteins changed directionality and significance for risk of mortality based on whether data was ANML-normalized, with 108 of the associations being significant in both, but opposite directionality. The paired Cohen D agreement between beta estimates was moderate (D = 0.61; 95% CI, 0.58-0.64).

In this cohort of critically ill patients, ANML-normalization attenuated correlations with immunoassays and largely changed the distribution and direction of significant mortality-associated proteins compared with non-ANML-normalization.