Exploring Anti-Aging Literature via ConvexTopics and Large Language Models

Area of Science:

• Biomedical Informatics
• Computational Biology
• Data Science

Background:

• The exponential growth of biomedical literature necessitates advanced methods for knowledge organization and trend identification.
• Existing clustering and topic modeling techniques (e.g., K-means, Latent Dirichlet Allocation [LDA]) often suffer from instability and lack of reproducibility due to sensitivity to initial parameters and local optima.

Purpose of the Study:

• To develop a scalable and interpretable clustering algorithm for analyzing large biomedical text datasets.
• To address the limitations of conventional methods by ensuring global optimality and stable topic generation.

Main Methods:

• A novel reformulation of a convex-optimization-based clustering algorithm was developed.
• The algorithm selects data exemplars to guarantee a global optimum, producing fine-grained and stable topics.
• The method was applied to approximately 12,000 PubMed articles related to aging and longevity.

Main Results:

• The proposed method successfully uncovered interpretable topics within the aging and longevity research domain.
• Expert validation confirmed the relevance of identified topics, which ranged from molecular mechanisms to lifestyle interventions (e.g., diet, exercise) and the gut microbiome.
• The algorithm demonstrated superior reproducibility and interpretability compared to K-means, LDA, and BERTopic.

Conclusions:

• This convex-optimization approach provides a robust and reproducible framework for topic modeling in large-scale biomedical literature.
• The method's ability to generate stable, interpretable topics facilitates knowledge discovery and trend analysis.
• This work lays the foundation for developing accessible tools for navigating and understanding complex biomedical research landscapes.