, one of the leading causes of vision loss in older adults. Researchers developed a novel machine learning approach to analyze retinal gene expression data, uncovering a more nuanced understanding of the role immune cells play in this neurodegenerative eye disease.){kind=link}
A recent study is shedding new light on the complex biology underlying Age-related Macular Degeneration (AMD), one of the leading causes of vision loss in older adults. Researchers developed a novel machine learning approach to analyze retinal gene expression data, uncovering a more nuanced understanding of the role immune cells play in this neurodegenerative eye disease.
Using transcriptome data from 453 donor retinas, 105 controls and 348 with AMD, the research team built an explainable ML pipeline that identified 81 genes capable of distinguishing AMD cases from healthy individuals. The model showed solid diagnostic performance (AUC-ROC of 0.80, CI: 0.70–0.92), suggesting these genes hold relevance in the biological changes that accompany the disease.
Notably, a majority of these genes showed enriched expression in retinal glial cells, especially microglia and astrocytes. This finding was further corroborated through cellular deconvolution analysis, which revealed consistent differences in glial cell composition between healthy and AMD-affected retinas.
While genome-wide association studies have previously implicated immune dysfunction in AMD, this study provides new insights at the molecular and cellular level. The researchers cross-referenced their gene expression findings with independent single-cell RNA data and AMD-GWAS datasets. This led to the identification of a novel genetic variant, rs4133124, at the PLCG2 locus. This variant was associated with AMD both in GWAS and expression quantitative trait loci analyses, adding weight to the case for PLCG2’s involvement in retinal immune regulation.
PLCG2, known for its role in immune signaling, has previously been implicated in autoimmune disorders and Alzheimer’s disease. The new findings position it as a potential player in the neuroinflammatory processes that characterize AMD.
The researchers took a critical stance toward conventional methods of differential gene expression analysis, which often rely on rigid statistical thresholds. Instead, their ML approach was able to capture complex, non-linear gene interactions, an important distinction given the natural variability of human gene expression and the intricacies of disease processes.
Co-expression network analyses revealed that 76% of the ML-identified genes are involved in immune response, extracellular matrix remodeling, and complement pathways, mechanisms long believed to be central to AMD. Interestingly, these networks differed between diseased and healthy tissues, suggesting a shift in cellular interactions as the disease progresses.
Access the full article published in nature genomic medicine here.