Expert Knowledge & Machine Understanding: Bridging Reactome's Ontology with LLM Semantic Embeddings

Susanna Bravi; Riccardo De Luca; Rosa Sicilia; Christine Nardini; Mario Santoro

cs.AIAug 28, 2026

Expert Knowledge & Machine Understanding: Bridging Reactome's Ontology with LLM Semantic Embeddings

Authors: Susanna Bravi, Riccardo De Luca, Rosa Sicilia, Christine Nardini, Mario Santoro

Organizations: Istituto per le Applicazioni del Calcolo Mauro Picone, Italian National Research Council, Rome, Italy. · Unit of Artificial Intelligence and Computer Systems, Department of Engineering, Universit`a Campus Bio-Medico di Roma, Rome, Italy · UniCamillus-Saint Camillus International University of Health Sciences, Rome, Italy

Abstract

Biological knowledgebases like Reactome provide high-quality pathways that include biological elements' relationships and textual descriptions (metadata). The quality of such pathways is granted by manual curation, that presents, however, significant scalability challenges. Lately, numerous NLP tools have been proposed to cope with this issue, leveraging textual information to automatically expand biological knowledgebases. However, little exploration has been done so far to assess whether relationships among textual descriptions mirror higher order biological relationships. This study explores whether human-written descriptions in Reactome can be used to infer the experts' defined global hierarchical structure. To test this, we extracted from Reactome the Homo Sapiens hierarchy of pathways and their reactions (Reactome Hierarchy), and used textual metadata to reconstruct a Semantic Hierarchy, combining a sentence transformer model (SPECTER2) with a modified agglomerative nesting algorithm and a graph reconstruction algorithm. Quantitative (Laplacian Spectral Distance and Bootstrapping) and qualitative (global topological metrics) analyses confirm our hypothesis and indicate that the global hierarchical structure of pathways can be inferred by experts textual metadata.

Figures & tables

Reactome ID	Type	Name	Description
R-HSA-9612973	TopLevelPathway	Autophagy	Autophagy is an intracellular degradation process that is triggered by cellular stresses. There are three primary types of autophagy - macroautophagy, chaperone-mediated autophagy (CMA) and late endosomal microautophagy.
R-HSA-9615710	Pathway	Late endosomal	Microautophagy (MI) is a non-selective autophagic pathway that involves internalisation of cytosolic cargo through invaginations of the lysosomal membrane.
R-HSA-1632852	Pathway	Macroautophagy	Macroautophagy (hereafter referred to as autophagy) acts as a buffer against starvation by liberating building materials and energy sources from cellular components.
R-HSA-9613829	Pathway	Chaperone Mediated Autophagy	In contrary to the vesicle-mediated macroautophagy, the chaperone mediated mechanism of autophagy selectively targets individual proteins to the lysosome for degradation.
R-HSA-9626253	Reaction	HSPA8 binds LAMP2a multimers	Intracellular proteins are targeted for proteolytic degradation in lysosome with the aid of chaperones. Heat shock cognate 71 kDa protein (HSPA8) transports substrates from the cytosol.
R-HSA-5682388	BlackBoxEvent	Autophagosome maturation	The mechanisms involved in the closure of the phagophore into an enclosed autophagosome are poorly understood.

Table 1: Sample of the description dataset . Dataset used to generate the sentence embeddings, concatenating Name and Description column.

Class / Subclass	Nodes	Class / Subclass	Nodes
Pathway (superclass)	2,807	ReactionLikeEvent (superclass)	0
CellLineagePath	12	BlackBoxEvent	2,894
TopLevelPathway	29	CellDevelopmentStep	22
		Depolymerisation	6
		FailedReaction	454
		Polymerisation	43

Table 2: Node composition of the Reactome Hyerarchy (RH) by ontological class. Distribution of nodes across the two primary ontological classes, Pathway and ReactionLikeEvent , and their respective subclasses in the Reactome Graph Database, filtered to retain only Homo sapiens -related nodes.

Figure 1: Laplacian spectral plots . Spectral density of the normalized graph Laplacian for three networks: (a) RH, (b) SH, (c) SH with all edges rewired. Densities are computed by convolving the eigenvalue spectrum with a Gaussian kernel ( $\sigma$ = 0.01) and normalizing by the number of eigenvalues. All panels share the same y-axis to allow direct comparison. The Jensen–Shannon-based Laplacian spectral distance between RH and each variant is 0.085 (SH) and 0.224 (SH fully rewired).

Bioinformatics

See all 41 papers

HypoKG: Evidence-Disciplined Biomedical Hypothesis Generation Beyond Endpoint Knowledge
Sep 10, 2026Dominic Okonkwo, Adetayo Okunoye, Ismailcem Budak ArpinarLLM Reasoning with Graphs Evidence-Grounded Reasoning
School of Computing University of Georgia
Language-encoded network topology enables large language models to reason about complex networks
Sep 3, 2026Ucchwas Talukder Utsha, Sakib Mostafa, James Zou +1Graph Representation Learning LLM Reasoning with Graphs
Department of Radiation Oncology, Stanford University, Stanford, California, USA · Department of Electrical Engineering, Stanford University, Stanford, California, USA · Department of Biomedical Data Science, Stanford University School of Medicine, Stanford, California, USA +1
TaxoBench: Can Deep Research Agents Retrieve and Organize? Evaluating the Synthesis Gap with Expert Taxonomies
Jan 18, 2026Ming Zhang, Jiabao Zhuang, Wenqing Jing +15AI Agent Benchmarks Deep Research Agents
Fudan University · Hunyuan Team, Tencent

Sep 10, 2026J/K move · Enter open · S save

HypoKG: Evidence-Disciplined Biomedical Hypothesis Generation Beyond Endpoint Knowledge

Dominic Okonkwo, Adetayo Okunoye, Ismailcem Budak Arpinar

School of Computing University of Georgia

LLM Reasoning with Graphs Evidence-Grounded Reasoning Scientific Hypothesis Generation Research Idea Generation Large Language Model-Guided Scientific Discovery Knowledge Graphs+1

Open paper AskPDF arXiv

Large language models (LLMs) can generate biomedical hypotheses, but it remains unclear whether they truly reason from scientific evidence or simply produce convincing-sounding ideas. To study this, we combine three major biological databases: the Kyoto Encyclopedia of Genes and Genomes (KEGG), Rhea, and UniProt, into a unified biochemical knowledge graph and construct a benchmark of 550 paths connecting enzyme sources to rare disease endpoints, yielding 13,200 hypotheses from six LLMs under four conditions varying the biological information each model receives: source enzyme only, full biological path, or source and disease endpoint only. Hypotheses are scored using an expert-derived five-criterion rubric on a 1-5 scale per criterion. We find that models given both the source and disease endpoint often produce the highest-scoring hypotheses, showing that LLMs can generate compelling ideas from minimal information. However, these hypotheses are less grounded in the evidence. In contrast, models given the full biological path generate hypotheses more consistent with known mechanistic relationships. We call this evidence-disciplined reasoning. To confirm this effect, we shuffled intermediate path steps while keeping endpoints fixed. Evidence grounding dropped significantly (delta = -0.793, p < 0.001), confirming models genuinely used path structure during reasoning. Our findings show that knowledge graphs support hypothesis generation in two ways: they identify biological endpoint pairs absent from the literature, and their mechanistic paths guide how LLMs reason between them.

Abstract

Bioinformatics

Explore similar work

HypoKG: Evidence-Disciplined Biomedical Hypothesis Generation Beyond Endpoint Knowledge

Language-encoded network topology enables large language models to reason about complex networks

TaxoBench: Can Deep Research Agents Retrieve and Organize? Evaluating the Synthesis Gap with Expert Taxonomies