Fig. 2
Serum peptidome biomarker model building with linear correlated features of worker CNT/F exposure. The 934 significantly responsive serum peptides were assessed by multivariate linear regression against metrics of CNT/F breathing zone levels, worker demographics and histories, and assay measures of serum inflammatory potential. a Out-of-sample predictive error was assessed by Akaike information criterion (AICc) scores for multivariate worker metric combinations. b Performance was best when taking the linear combination of three personal breathing zone CNT/F measures, log∑(Exp), with linearity exemplified for peptide KELGALQQC. c Moderate correlation (Shapiro R2) was observed across 39% of the responsive peptides for respirable EC measures (logResEC), with performance improved when adding inhalable EC (logInhResEC) and CNT/F structure counts (log∑(Exp)). Additional serum peptidome variability was explained by covariates for serum induction of IL6 and worker age. d-e Venn diagrams illustrating the overlap of strongly correlated (R2 > 0.7) serum peptides between multivariate worker metrics