AbstractChildren develop rapidly during the first years of life, and understanding the sources and associated levels of variation in the serum proteome is important when using serum proteins as markers for childhood diseases. The aim of this study was to establish a reference model for the evolution of a healthy serum proteome during early childhood. Label-free quantitative proteomics analyses were performed for 103 longitudinal serum samples collected from 15 children at birth and between the ages of 3–36 months. A flexible Gaussian process-based probabilistic modelling framework was developed to evaluate the effects of different variables, including age, living environment and individual variation, on the longitudinal expression profiles of 266 reliably identified and quantified serum proteins. Age was the most dominant factor influencing approximately half of the studied proteins, and the most prominent age-associated changes were observed already during the first year of life. High inter-individual variability was also observed for multiple proteins. These data provide important details on the maturing serum proteome during early life, and evaluate how patterns detected in cord blood are conserved in the first years of life. Additionally, our novel modelling approach provides a statistical framework to detect associations between covariates and non-linear time series data.