DisMech Use Cases

This document outlines use cases for the Disorder Mechanisms Knowledge Base (DisMech), organized by audience and application domain.

Clinical & Research

Differential Diagnosis Support

Given a set of phenotypes (HP terms), query DisMech to find disorders sharing those phenotypes and rank by overlap. The structured phenotype descriptors with ontology bindings make this computationally tractable.

Mechanism-Based Drug Repurposing

Two diseases that share pathophysiology (e.g., same biological processes, same cell types affected) but have different treatments could be candidates for cross-disease drug repurposing. The structured pathophysiology, cell_types, and treatments fields enable systematic comparison.

Comorbidity Prediction

The comorbidities and disease trajectory data can be mined to identify unexpected disease co-occurrences and their shared mechanistic basis, going beyond epidemiological correlation to mechanistic explanation.

Clinical Trial Matching

With clinical_trials, phenotypes, and genetic_basis structured and ontology-linked, patients could be matched to relevant trials based on their genotype-phenotype profile.

Education

Interactive Pathophysiology Browser

The HTML rendering already provides a browsable view of each disorder. This could be expanded into a teaching tool where students explore how genetic variants lead to molecular dysfunction, cellular changes, and clinical phenotypes — the full "mechanism chain."

Evidence Literacy Training

The evidence model (SUPPORT/REFUTE/PARTIAL, with required PMID snippets) teaches trainees to evaluate claims against primary literature rather than accepting statements at face value.

Bioinformatics & Data Integration

Ontology-Grounded NLP Benchmark

DisMech entries with their precise ontology mappings (HP, MONDO, GO, CL, MAXO) could serve as gold-standard annotations for evaluating biomedical NLP systems that extract disease mechanisms from text.

Knowledge Graph Seeding

Each disorder file encodes a mini knowledge graph (disease → genetic basis → pathophysiology → cell types → phenotypes → treatments). These could be projected into a formal KG (e.g., Monarch Initiative's KG) to enrich disease-gene-phenotype edges with mechanistic context.

Cross-Ontology Bridging

DisMech links MONDO diseases to HP phenotypes, GO processes, CL cell types, MAXO treatments, and UBERON anatomy in a single curated record. This creates implicit cross-ontology mappings that are otherwise hard to derive automatically.

Microbiome Integration

Diseases with environmental or microbial components (e.g., infectious diseases, gut-related conditions) could be enriched with microbiome data from resources like NMDC biosamples, linking host disease mechanisms to microbial community profiles.

AI & Curation

LLM Hallucination Benchmarking

The reference validation pipeline (snippet matching against PubMed abstracts) is itself a reusable pattern for measuring how accurately LLMs cite scientific literature. DisMech could publish a benchmark dataset of verified vs. hallucinated citations.

Automated Curation Pipelines

The existing skill system demonstrates a pattern where AI agents curate structured knowledge with human-in-the-loop validation. This pattern could generalize to other biomedical knowledge bases.

Compliance-Driven Prioritization

The weighted compliance scoring identifies which entries most need enrichment. This drives a triage system where curator effort (human or AI) is directed to the highest-impact gaps.

Precision Medicine

Genotype-to-Treatment Pathway Mapping

For entries with genetic_basis → pathophysiology → treatments chains (especially cancer entries like BRAF V600E melanoma, ALK-rearranged NSCLC), DisMech encodes the logic of precision oncology: specific mutations → specific targeted therapies.

Rare Disease Mechanism Cataloging

With disorders including rare conditions (Achondrogenesis Type II, Beta-Mannosidosis, CHIME syndrome), DisMech serves as a structured mechanism reference for diseases that are often poorly documented in traditional resources.