Graph AI generates neurological hypotheses validated in molecular, organoid, and clinical systems

2025-12-13

AI Summary

Neurological disorders impose a substantial global disability burden, yet effective disease-modifying therapies remain scarce. To address this, we propose PROTON—the first heterogeneous graph Transformer model designed for Parkinson’s disease (PD), bipolar disorder (BD), and Alzheimer’s disease (AD). It integrates multi-scale data—including genomics, organoid assays, and clinical records—to establish a novel “AI-driven hypothesis generation → multi-level collaborative validation → causal inference” paradigm. Methodologically, PROTON unifies heterogeneous graph neural networks, graph attention mechanisms, multimodal knowledge graph embeddings, and real-world electronic health record–based causal analysis. Experimentally, it identifies neurotoxic pesticides (e.g., endosulfan, top 1.29%) in PD with high precision and replicates six α-synuclein experimental findings (normalized enrichment score >2.13, FDR <1×10⁻⁴); discovers calcitriol as a candidate pathological protein modulator in BD; and predicts five AD therapeutics significantly reducing 7-year dementia incidence (hazard ratio = 0.63, p <1×10⁻⁷).

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Relational Probabilistic Models
• Humans and AI: Other Foundations of Human Computation & AI

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

Neurological diseases are the leading global cause of disability, yet most lack disease-modifying treatments. We present PROTON, a heterogeneous graph transformer that generates testable hypotheses across molecular, organoid, and clinical systems. To evaluate PROTON, we apply it to Parkinson's disease (PD), bipolar disorder (BD), and Alzheimer's disease (AD). In PD, PROTON linked genetic risk loci to genes essential for dopaminergic neuron survival and predicted pesticides toxic to patient-derived neurons, including the insecticide endosulfan, which ranked within the top 1.29% of predictions. In silico screens performed by PROTON reproduced six genome-wide $α$-synuclein experiments, including a split-ubiquitin yeast two-hybrid system (normalized enrichment score [NES] = 2.30, FDR-adjusted $p < 1 imes 10^{-4}$), an ascorbate peroxidase proximity labeling assay (NES = 2.16, FDR $< 1 imes 10^{-4}$), and a high-depth targeted exome sequencing study in 496 synucleinopathy patients (NES = 2.13, FDR $< 1 imes 10^{-4}$). In BD, PROTON predicted calcitriol as a candidate drug that reversed proteomic alterations observed in cortical organoids derived from BD patients. In AD, we evaluated PROTON predictions in health records from $n = 610,524$ patients at Mass General Brigham, confirming that five PROTON-predicted drugs were associated with reduced seven-year dementia risk (minimum hazard ratio = 0.63, 95% CI: 0.53-0.75, $p < 1 imes 10^{-7}$). PROTON generated neurological hypotheses that were evaluated across molecular, organoid, and clinical systems, defining a path for AI-driven discovery in neurological disease.

Problem

• PROTON generates testable hypotheses for neurological diseases across multiple systems.
• It links genetic risk to molecular mechanisms and predicts therapeutic candidates.
• The model validates predictions in clinical data to reduce dementia risk.

Innovation

• PROTON uses heterogeneous graph transformer for neurological hypotheses
• It predicts drug candidates and toxic substances across biological systems
• Validates predictions via molecular, organoid, and clinical data integration