At the 27th Annual Conference of the Chinese Society of Diabetes (CDS 2025), Professor Chen Yanming from the Third Affiliated Hospital of Sun Yat-Sen University delivered a presentation on multi-omics-based prediction and AI-powered precision diagnosis of diabetic retinopathy (DR).
Multi-Omics-Based Prediction and AI-Powered Precision Diagnosis of DR
The application of multi-omics technologies — including genomics, epigenomics, transcriptomics, proteomics, and metabolomics — allows the identification of molecular-level metabolic imbalances in DR before the onset of structural lesions.
Metabolomics studies have revealed that changes in retinal-related metabolites, such as those in the leucine/isoleucine pathway and lipid oxidation products, precede vascular injury and can serve as early metabolic warning signals.
Proteomic profiling has identified molecular markers of DR including ANGPTL4 and PPAR-related proteins, and further screened blood-based indicators such as PLXNB2, GDF15, and REN, which significantly improve the prediction of DR onset and progression.
In addition, a panel of multi-dimensional network metabolites (including linoleic acid, nicotinic acid, ornithine, and phenylacetylglutamine) can effectively distinguish diabetic patients with DR from those without DR, achieving a sensitivity of 96% and specificity of 78%.
Reduced circulating L‑tyrosine levels indicate DR risk and early neurodegeneration, with a predictive sensitivity of 86% and specificity of 40%.
Based on large-scale single-nucleus RNA and ATAC sequencing data, the human retinal single-cell multi-omics atlas has constructed regulatory networks for more than 200 transcription factors. This enables the precise mapping of genome-wide association study (GWAS) risk loci and proteomic markers to specific cell types, providing a foundation for DR classification and individualized intervention.
Meanwhile, artificial intelligence (AI) technology is evolving from single-modal imaging diagnosis to the integration of multi-modal metabolic indicators, enabling the construction of more sensitive risk prediction models. Systems such as ARDA, RetinAI, and the domestic platform SELENA+ are now at the forefront of clinical application.