Diabetes mellitus (DM) is a chronic metabolic disease caused by multiple etiologies.Diabetic retinopathy (DR), as a primary ocular complication of DM, is the leading cause of blindness among working-age adults in the world.With the rapid increase of diabetes incidence worldwide, the diagnosis of DR is often delayed because few symptoms of the retinal vessel-nurse unit lesion are found in early DR.Therefore, the early diagnosis, prevention and treatment of DR are facing much more challenges.At present, the early clinical biomarkers of DR, such as glycosylated hemoglobin and blood glucose, are of great value in predicting and preventing the occurrence and development of DR, but there is still a lack of research on the pathological effect of metabolic heterogeneity and its potential induction of DR.Multi-omics methods, such as metabolomics and single-cell transcriptome, as well as deep learning techniques, are powerful tools for the study of DR pathophysiological processes, which can be used to reveal the metabolic characteristics of DR, discover early biomarkers and new metabolic pathways and identify targets for treatment.Future advances which aim to diagnose and treat DR should consider the metabolic remodeling induced by genetic background and environmental factors comprehensively, combine omics approaches and the measurement of clinical indicators of DR occurrence and development to find biomarkers of early DR and targets so as to achieve early prediction and accurate prevention of DR.