Predictive biomarkers for acute graft-versus-host disease (aGVHD) is currently lacking. In this study, we employed an unbiased proteome profiling method to prospectively collected plasma samples from allogeneic hematopoietic stem cell transplantation (alloHSCT) recipients to identify protein biomarkers that predict the risk of aGVHD and non-relapse mortality (NRM). In the discovery set, including five aGVHD patients and five controls, we identified seven candidate proteins. Patients with high levels of these proteins tended to exhibit a higher risk of aGVHD and NRM compared to patients with low levels in post-engraftment plasma samples from an independent validation set (n = 89). Tissue inhibitor of metalloproteinase 1, plastin-2, and regenerating islet-derived protein 3-α were selected as the most-predictive biomarkers via an exhaustive variable screening algorithm and were collectively used to develop a biomarker panel score ranging from 0 to 3. The biomarker panel score correlated significantly with aGVHD and NRM risk in univariable and multivariable Cox models. Furthermore, using the biomarker panel score in conjunction with clinical predictors significantly improved the discriminatory performance of the Cox model in predicting aGVHD and NRM risk. Our findings suggest that plasma-derived protein biomarkers can be used to predict aGVHD and NRM before the onset of clinical manifestations.