Locally advanced esophageal squamous cell carcinoma (ESCC), often treated with concurrent chemoradiotherapy (CCRT) to improve control rate and survival with little reliability in efficacy due to greater hematologic toxicity in patients. Examining serum metabolomics and proteomics, Wu et al. analyzed and created prediction models for prognostic and hematologic toxicity. Machine learning integrated GC-MS metabolomics and Olink proteomics to predict treatment response, survival, and hematologic toxicity in ESCC patients 3 months post-CCRT, with siRNA transfection experiments confirming the SHMT2-serine pathway mechanism. This model used four serum metabolites and nine proteomic proteins to highly accurately predict CCRT efficacy with complete vs noncomplete response and biomarkers classifying biomarkers for patient responses. Biologically, the researchers demonstrated elevated L-serine in high-risk patients often correlated led to poor survival and non-complete treatment response; in comparison, normalized L-serine levels post-treatment often led to significantly better patient outcomes. This was validated by the serine metabolism pathway and demonstrated treatment response is dependent on extracellular L-serine levels. In conclusion, Wu et al. formulated a predictive model for treatment efficacy and survival often as a joint action of metabolic and immune function. Lower serum L-serine was linked to better prognosis and treatment sensitivity effect dependent on serine availability. This integration of Olink proteomics with metabolomics improved biomarker discovery, allowing for multiomics approaches for prognosis prediction.