PRISMA: A tensor-based framework for deconstructing the genetic architecture of complex diseases, with application to diabetic retinopathy

The paper introduces PRISMA, a novel tensor-based framework that deconstructs complex-disease GWAS signals into tissue-resolved genetic trajectories by integrating summary statistics with multi-tissue eQTL data, successfully revealing distinct vascular, immune, and neurodegenerative axes in diabetic retinopathy that traditional methods fail to capture.

The Gist

Imagine that trying to understand a complex disease like diabetic retinopathy using standard genetic studies is like listening to a massive, chaotic orchestra playing a single, loud chord. You can hear the noise (the disease risk), but you can't tell which instruments are playing, which section is out of tune, or how the different parts are working together to create that sound.

Standard genetic tests usually give you a "global" score for a specific spot in your DNA. It's like saying, "This section of the orchestra is loud," without telling you if it's the violins, the drums, or the brass causing the noise. This hides the specific details of how different body tissues contribute to the disease.

Enter PRISMA: The Genetic "Sound Engineer"

The paper introduces a new tool called PRISMA. Think of PRISMA as a sophisticated sound engineer who can take that messy, loud chord and break it down into its individual instrument tracks.

Here is how it works, using simple metaphors:

• The Problem: Standard tests mix up signals from different body parts (like the eyes, the blood vessels, and the immune system) into one big pile.
• The Solution: PRISMA uses a special mathematical technique (described as "graph Laplacian-regularized block-wise factorization") to untangle that pile. It looks at how genes are connected in specific tissues, preserving the natural "neighborhood" of DNA (called linkage disequilibrium) so it doesn't accidentally mix up signals that belong to different areas.
• The Result: Instead of one big lump of risk, PRISMA separates the genetic risk into three distinct "trajectories" or paths:
  1. Vascular-Metabolic: Issues related to blood vessels and sugar metabolism.
  2. Systemic Immune-Inflammatory: Issues related to the body-wide immune system and inflammation.
  3. Retina-Specific Neurodegenerative: Issues specific to the nerve cells in the eye itself.

What Did They Find?

When the researchers applied this tool to diabetic retinopathy, they didn't just find the usual suspects. They uncovered 549 specific genetic targets that drive these different paths.

• The "Hidden Gems": Many of these targets (403 of them) were previously invisible to standard tests because they didn't meet the strict "genome-wide significance" threshold. PRISMA found them by looking at the specific tissue context, much like finding a quiet soloist in a loud room by turning up the volume on just that instrument.
• Better than the Old Tools: The paper claims PRISMA does a better job of separating these tissue-specific signals than other common methods like PCA, NMF, or K-means (which are like trying to sort a mixed bag of marbles by color using a very rough sieve).
• Proof it Works:
  • They tested it on a "height" study (a control group) to show the method can distinguish between traits that share genetic maps but act differently in different tissues.
  • They checked their findings against single-cell data (looking at individual cells) and found that the three paths they identified matched up perfectly with specific cell types: fibrovascular cells, immune cells, and retinal cells.
  • They even looked at proteins and metabolites in the "vitreous humor" (the jelly inside the eye) and found molecules that matched these genetic paths.

The Bottom Line

The paper argues that we need to stop looking at complex diseases as a single, blurry block of genetic risk. Instead, PRISMA allows scientists to map out the specific, tissue-by-tissue routes that lead to disease. It turns a blurry, aggregate picture into a sharp, high-definition map of how different parts of the body contribute to the problem.

Technical Summary

Technical Summary of PRISMA: A Tensor-Based Framework for Deconstructing Genetic Architecture

Problem Statement

Current Genome-Wide Association Studies (GWAS) for complex diseases typically compress heterogeneous, tissue-dependent genetic effects into global, locus-level summary statistics. This aggregation masks the specific local regulatory routes through which polygenic risk contributes to disease architecture. Consequently, the distinct tissue-specific mechanisms driving disease progression remain obscured within aggregate signals, limiting the ability to identify precise therapeutic targets and understand the biological trajectories of complex traits.

Methodology

To address these limitations, the authors introduce PRISMA (Polygenic Risk Integration via Summary-statistics Multi-tissue Array-decomposition), a novel summary-statistics framework designed to quantify tissue-resolved genetic heterogeneity. The core technical approach involves:

• Data Integration: PRISMA integrates GWAS summary statistics with multi-tissue cis-expression Quantitative Trait Loci (eQTL) data.
• Tensor Decomposition: The framework employs graph Laplacian-regularized block-wise factorization. This mathematical approach decomposes the integrated data tensor while explicitly preserving the local Linkage Disequilibrium (LD) topology during the decomposition process.
• Comparative Validation: The method's performance is benchmarked against standard dimensionality reduction and clustering techniques, specifically Principal Component Analysis (PCA), Non-negative Matrix Factorization (NMF), and K-means clustering.
• Control Analysis: A negative-control analysis using a height GWAS was conducted to validate the framework's ability to distinguish trait-dependent regulatory reprioritization within a shared eQTL-mappable context.
• Multi-Omics Validation: The identified axes were further validated using independent single-cell transcriptomic data, as well as exploratory proteomic and metabolomic profiling of vitreous humor.

Key Results

Applied to diabetic retinopathy, PRISMA successfully deconvolved aggregated polygenic risk into three distinct, tissue-biased genetic axes:

1. Vascular-metabolic trajectory
2. Systemic immune-inflammatory trajectory
3. Retina-specific neurodegenerative trajectory

Key quantitative findings include:

• Target Prioritization: The framework prioritized 549 axis-associated targets. Notably, 403 of these targets fell below the conventional genome-wide significance threshold, suggesting that PRISMA can recover biologically relevant signals often missed by standard GWAS filtering.
• Resolution Superiority: PRISMA demonstrated higher tissue-regulatory resolution compared to PCA, NMF, and K-means.
• Biological Concordance: Independent single-cell transcriptomic analyses confirmed axis-specific enrichment across fibrovascular, immune, and retinal compartments. Furthermore, exploratory vitreous humor proteomic and metabolomic profiling nominated candidate molecular correlates representing downstream convergence of these genetic trajectories.

Significance and Claims

The paper posits that PRISMA represents a paradigm shift in the analysis of complex-disease GWAS. Rather than focusing solely on aggregate locus discovery, the framework enables quantitative, LD-aware mapping of tissue-resolved genetic trajectories. By deconstructing heterogeneous genetic effects into specific regulatory routes, PRISMA offers a more granular understanding of disease architecture, facilitating the identification of tissue-specific therapeutic targets that are otherwise obscured in global statistical summaries. The authors frame this as a move toward a more precise, mechanism-driven interpretation of polygenic risk.