Bridging Bibliometrics and Biology: A Framework for the Relevance Scoring and Biological Stratification of Scalable MeSH Co-occurrence Networks

Systematic literature review is often limited by the rigidity of search strings, which often miss non-obvious but highly relevant topics. Controlled vocabularies such as Medical Subject Headings (MeSH) facilitate the review process, but relationships between terms, and their relevance to a hypothesis of interest, are rarely stated explicitly. Presented here is a MeSH term extraction and networking technique, seeded by a single query for Allergic Contact Dermatitis, forming a "themed" network of terms. The resulting co-occurrence network is pruned by an unsupervised consensus of Metropolis-criterion likelihood search strategies, isolating a significant core subgraph and filtering out incidental co-indexing. Symmetrically normalized Article Relevancy Scores (ARS) then evaluate each article against the network's weighted terms across the annotated MEDLINE corpus, correcting for publication volume bias; Mean Relevancy Scoring (MRS) inverts these scores back onto the terms, scoring each MeSH term by the mean ARS of the articles carrying it to measure its specificity to the query. Benchmarked against the expert-curated Adverse Outcome Pathway (AOP) for Allergic Contact Dermatitis, the network recovered key event terminology and identified highly specific literature that standard retrieval methods overlooked. This data-driven approach quantitatively assesses associated MeSH terminology, extracting biologically relevant concepts and key relationships directly from the underlying structure of the literature. By automating the discovery of these latent connections, the framework provides a reproducible, scalable foundation for understanding the biomedical literary space surrounding a given search string, assisting systematic review creation, mechanistic pathway mapping, and hypothesis generation across biomedical disciplines.