Artificial Intelligence-Based Thematic Analysis of Biomedical Literature for Precision Nutrition

Area of Science:

• Biomedical Informatics
• Computational Biology
• Bibliometrics

Background:

• Contemporary biomedical research, especially in precision nutrition, faces challenges in maintaining comprehensive understanding due to the exponentially increasing volume of scientific literature.
• Traditional literature reviews struggle to capture the structural landscape of entire research fields, hindering the integration of diverse data streams crucial for understanding metabolic responses.
• The growing body of research diminishes visibility across the full evidence landscape, impacting feature selection and model interpretability in fields like precision nutrition.

Purpose of the Study:

• To develop and validate an end-to-end computational framework for organizing and analyzing large-scale biomedical literature.
• To assess the representativeness of traditional reviews by comparing their thematic structures to those derived from primary research abstracts.
• To provide a scalable solution for identifying knowledge gaps and supporting evidence-grounded modeling in precision nutrition research.

Main Methods:

• An automated framework was created to ingest PubMed query results and apply unsupervised thematic structuring to abstracts.
• Natural language processing techniques were employed for abstract analysis, followed by independent clustering of review and primary research subsets.
• Cluster-centroid cosine similarity heat maps were used to compare thematic distributions, and the framework's scalability was demonstrated with over 385,000 abstracts.

Main Results:

• The framework successfully mapped over 385,000 abstracts related to genetics and physical activity into stable thematic representations.
• A strong correspondence was observed between review and primary research clusters in clinically consolidated domains.
• Mechanistic and genomics-focused research areas showed comparatively limited representation within traditional review-derived structures, indicating potential review gaps.

Conclusions:

• The developed computational framework offers a scalable approach to structural analysis of biomedical literature, transforming literature synthesis.
• This tool provides quantitative infrastructure for evidence-grounded modeling, particularly beneficial for the complex demands of precision nutrition research.
• An interactive web application operationalizes the pipeline, enabling reproducible corpus-scale mapping and review-gap diagnostics without manual intervention.