Abstract
The field of biomedical science is one of the most studied areas of the 21st century, the field has published a huge number of research papers, which have averaged more than 600,000 articles a year. How to effectively obtain knowledge in the vast literature of research is a challenge for researchers in the field. As one of the branches of bioinformatics, technology of text mining of biomedical field is a new exploration of the efficient and automated acquisition of relevant knowledge. In this paper, we take the example of genetic enhancement, articles related to Genetic Enhancement from 2005 to 2016 were selected as datasets. We focus on two approaches, co-word analysis is used to identify research hotspot and complex network analysis is used to analyze the collaboration network to determine the status of research collaboration in this field. From co-word analysis, we find that there are four research hotspots, they are gene expression related field, bioethical issues related field, metabolic and protein engineering related area and cell related field. From research collaboration analysis, we find that the research collaboration network has a scale-free feature and the max connected subgraph of it has a small world phenomenon. We also find that research collaboration in this field is now recovering.
Introduction
The biomedical field is currently experiencing an unprecedented data explosion, with an average of over 600,000 new articles published annually. For researchers, staying abreast of "implicit knowledge"—the hidden connections between topics or the structural evolution of a field—is nearly impossible using traditional keyword searches. This paper addresses this information overload by applying automated text mining to synthesize large volumes of literature, specifically targeting the field of Genetic Enhancement to demonstrate how computational tools can identify research trends and evaluate the health of scientific collaboration.
Methodology
The researchers employed two primary computational strategies to analyze 1,380 articles from the PubMed/MEDLINE database published between 2005 and 2016:
- Co-word Analysis: This technique identifies "hotspots" by analyzing the co-occurrence of Medical Subject Headings (MeSH) terms. The study used Zipf’s Law to establish a threshold for high-frequency words and applied Ochiai similarity for correlation, followed by hierarchical clustering.
- Complex Network Analysis: To understand the social structure of the field, a collaboration network was constructed where nodes represent authors and edges represent co-authored papers. Metrics such as degree distribution, clustering coefficients, and K-core analysis were used to evaluate the network's topology.
Mapping Research Hotspots
Using Zipf's Law, the authors identified 67 high-frequency MeSH words (appearing more than 78 times) that define the field's core interests. The subsequent hierarchical clustering revealed that the field is organized into four distinct research pillars.
These four clusters include:
- Gene Expression: Focused on the fundamental biological mechanisms.
- Bioethical Issues: Highlighting that social and legal concerns are as central to the field as the science itself.
- Metabolic and Protein Engineering: Focusing on the practical applications of genetic modification.
- Cell-Related Research: Investigating the cellular foundations of enhancement technologies.
The Social Structure of Science
The collaboration network analysis revealed a "Small World" phenomenon. With an average shortest path of only 2.27, any two researchers in this global field are connected by just over two "handshakes," facilitating rapid information flow.
Key findings regarding the network include:
- Scale-Free Dynamics: The network follows a power-law distribution, meaning a few highly connected "hubs" drive the field while most researchers have few connections.
- High Local Clustering: The average clustering coefficient in the max connected subgraph is 0.93, indicating extremely tight-knit research groups.
- Core vs. Periphery: K-core analysis showed that 90% of authors belong to the network periphery, suggesting a small elite drives the majority of collaborative efforts.
- Collaboration Recovery: Data indicates that both the number of authors per paper and the percentage of co-authored papers began to rise significantly after 2011, suggesting a shift toward more interdisciplinary team science.
Conclusion
This study demonstrates that text mining and complex network analysis can effectively map the intellectual and social landscape of a complex biomedical field. By identifying core research hotspots and the "small-world" nature of scientific collaboration, these tools provide researchers and funding agencies with a high-level structural overview that traditional literature reviews cannot achieve. Future work may involve using these quantitative indexes to identify underserved areas and emerging subject forces more precisely.
Summary updated