OMICS-BASED PROFILING IN INHERITED RETINAL DYSTROPHIES: DISSECTING OLIGOGENIC INHERITANCE THROUGH MUTATIONAL BURDEN MODELS AND DEVELOPING A PLASMA PROTEOME-BASED FRAMEWORK

Inherited Retinal Dystrophies (IRDs) are a genetically and clinically heterogeneous group of disorders leading to progressive and irreversible vision loss. Although over 250 genes have been associated with IRDs, approximately 40% of patients remain genetically unresolved after next-generation sequencing. In many cases, only single heterozygous variants in recessive genes are identified, suggesting the involvement of complex inheritance models such as oligogenicity. The concept of mutational burden, which captures the cumulative effect of rare variants across functionally related genes, represents a promising approach to interpret these cases. In parallel, IRD progression is not linear but characterized by alternating latent and active phases, without reliable predictors. Plasma proteomics offers an opportunity to identify molecular signatures associated with disease activity, including inflammation, oxidative stress, and cell death. This study addresses both the genetic and temporal complexity of IRDs through two complementary approaches. First, we developed a mutational burden model to quantify the aggregate impact of rare heterozygous variants across retinal pathways. Second, we implemented a machine-learning classifier based on plasma proteomic data to distinguish between active and latent phases of disease. We analyzed data from 527 individuals (453 IRD patients and 74 controls). Variants were classified according to ACMG guidelines and assigned weighted pathogenicity scores, normalized by age. Pathway-based mutational burden was calculated using curated databases (KEGG, Reactome, STRING), and statistical differences were assessed. High-burden patients were identified using a 90th percentile threshold, and gene combinations unique to patients were explored. The model identified significant enrichment in retinoid cycle pathways and highlighted gene pairs (ABCA4–RDH12 and ABCA4–RBP3) present exclusively in patients, supporting potential oligogenic mechanisms. In parallel, a synthetic dataset of 500 plasma proteomic profiles (315 proteins) was used to train machine-learning models. Logistic Regression showed the best performance (accuracy 72%, F1-score 0.71), identifying key biomarkers including TIMP3, VEGFA, CAPN5, CLN3, and PRDM13. These approaches improve variant interpretation, patient stratification, and disease monitoring. The mutational burden model supports reclassification of unresolved cases beyond Mendelian frameworks, while the proteomic classifier enables detection of disease activity phases. Together, they provide a scalable, data-driven framework for precision medicine in IRDs, with potential applications in diagnostics, clinical decision support, and trial stratification. Future work will focus on validation in independent cohorts and integration with clinical and imaging data.