DAG & RAG for Amyotrophic Lateral Sclerosis Diagnostics

This repository contains the implementation of our study titled "Bridging the Omic Tipping Point: Vector Knowledge Retrieval Unmasks Individual Causal Mechanics in Amyotrophic Lateral Sclerosis Diagnostics".

Authors

• Ali Salman - Department of Medical Biotechnologies, University of Siena, 53100 Siena, Italy.
• Alessio Rotelli - Department of Medical Biotechnologies, University of Siena, 53100 Siena, Italy.
• Matteo Leoncini - Department of Medical Biotechnologies, University of Siena, 53100 Siena, Italy.
• Leya El Halabi - Department of Medical Biotechnologies, University of Siena, 53100 Siena, Italy.
• Forough Izadi - Department of Medical Biotechnologies, University of Siena, 53100 Siena, Italy.
• Elena Niccolai - Department of Clinical and Experimental Medicine, University of Florence, Florence, Italy.
• Jessica Mandrioli - Department of Clinical and Experimental Medicine, University of Florence, Florence, Italy.
• Amedeo Amedei - Department of Clinical and Experimental Medicine, University of Florence, Florence, Italy.
• Ernesto Iadanza - Department of Medical Biotechnologies, University of Siena, 53100 Siena, Italy.

Abstract

Amyotrophic Lateral Sclerosis (ALS) is a devastating neurodegenerative disease with complex, poorly understood molecular mechanisms. Traditional machine learning diagnostics function as opaque ``black boxes,'' providing predictions based on correlation without clinical or structural accountability. This study introduces a novel diagnostic paradigm that replaces traditional predictive modeling with a framework built on structural causal graph metadata stored in a Retrieval-Augmented Generation (RAG) system. Using a clinical cohort of 118 individuals (97 ALS patients, 21 healthy controls), we applied constraint-based causal discovery (the PC Algorithm) to map direct cause-and-effect pathways across 34 continuous blood and fecal biomarkers. Rather than storing raw data vectors, the resulting network's structural metadata (specifically directional causal arrows and edge coefficients) was embedded into a permanent vector database. Forward clinical inference was achieved entirely by querying this RAG system to dynamically trace patient biomarker values along the graph's metadata pathways, calculating a diagnostic probability using a robust, matching-sign algebraic framework. The causal discovery pipeline successfully mapped a localized biological network of 45 verified causal pathways, revealing explicit disease-aligned mechanisms such as the downstream suppression of Interleukin-2 (IL-2). By computing patient risk profiles exclusively through structural metadata traversal, the RAG predictor achieved an elite global accuracy of 99.15%, a clinical sensitivity of 98.97%, and a flawless specificity of 100.00%. The novelty of this architecture lies in its ability to extract forward-looking diagnostic inferences directly from stored causal metadata, entirely eliminating the need for traditional machine learning classification layers. This framework bridges the gap between high-accuracy prediction and absolute clinical transparency, providing automated, patient-specific causal audits that isolate precise metabolic risk disruptors and protective mechanisms for targeted therapeutic discovery.

Key Features

• Hybrid Causal-RAG Architecture: Combines structural equation modeling (PC algorithm) with vector retrieval (ChromaDB) to ground data-driven causal discoveries in biomedical literature.
• Patient-Specific Mechanics: Moves beyond population-level averages to unmask individualized causal pathway dynamics across multi-omic profiles in ALS.
• Latent Space Knowledge Mapping: Projects structural path weights (β) and textual narrative embeddings into a shared vector space, clustering inflammatory cytokines and metabolomic markers.
• Non-Negative Causal Boundary Constraints: Applies explicit clipping mechanisms (max(0,x)) to prevent biologically implausible negative path values in directional pathways.
• Fully Reproducible Pipeline: Built on an open-source, lightweight stack using causal-learn, chromadb, and modular Python scripts designed for seamless execution.

Citation

If you use this work, please cite it as:

@misc{Salman2026,
  author       = {Ali Salman and Alessio Rotelli and Matteo Leoncini and Leya El Halabi and Forough Izadi and Elena Niccolai and Jessica Mandrioli and Amedeo Amedei and Ernesto Iadanza},
  title        = {Bridging the Omic Tipping Point: Vector Knowledge Retrieval Unmasks Individual Causal Mechanics in Amyotrophic Lateral Sclerosis Diagnostics},
  year         = {2026},
  institution  = {Department of Medical Biotechnologies - University of Siena, Siena, Italy},
  url          = {https://github.com/alexsalman/als/}
  urldate      = {2026-08-01}
}

⸻
© 2026 Ali Salman · MIT License