Construction of Viral Hepatitis Bilingual Bibliographic Database with Protein Text Mining and Information Integration Functions

Heng Chen, Yongjuan Zhang, Chunhong Lin, Liwen Zhang, Tao ChenSML @ IJCAI 2017

Abstract

Viral hepatitis research is characterized by a high volume of bilingual literature and a disconnect between bibliographic data and functional protein information. This paper details the construction of a specialized database that integrates Medline and CNKI literature with the UniProt protein resource. By employing a three-stage indexing process and a bilingual protein dictionary, the system enables automated text mining and direct linking between research papers and protein analysis tools, facilitating a more integrated workflow for life science researchers.

Introduction

The field of life sciences currently faces a "data-rich, knowledge-poor" paradox. In viral hepatitis research, thousands of papers are published annually in both English and Chinese, yet this information often remains siloed from factual protein databases. Researchers frequently encounter friction when moving from literature review to sequence analysis. This study presents a unified platform designed to bridge these gaps through automated bilingual text mining and functional integration, allowing for a "one-stop" research experience.

Methodology and System Architecture

The platform is built on a 3-tier architecture (Client, Web Server, Database Server) using a Linux/Nginx/MySQL/PHP/C++ stack. The integration pipeline is the system's core, facilitating the flow from raw literature to an indexed, searchable knowledge base.

Figure 2: The data integration pipeline, illustrating the flow from bilingual literature sources and protein dictionaries to the indexed subject database.

Key components of the methodology include:

  • Bilingual Mapping: A specialized dictionary translates Chinese protein terms (e.g., "乙型肝炎 e 抗原") into English equivalents ("HBeAg") to facilitate cross-lingual mining.
  • Quality-Controlled Indexing: A three-stage process involving preprocessing, dictionary-based control, and post-indexing feedback ensures high data quality and minimizes false positives.
  • Protein Text Mining: C++ modules identify protein mentions in text and map them to UniProt entries, creating a direct link between literature and factual data.

Data Integration

The system draws from three primary repositories to ensure comprehensive coverage:

  1. Medline: The gold standard for English-language biomedical literature.
  2. CNKI (China National Knowledge Infrastructure): The primary source for Chinese-language academic datasets.
  3. UniProt: The authoritative resource for factual protein sequence and functional data.

Results and User Experience

The resulting database demonstrates significant technical performance, supporting up to 200 million records with sub-0.03-second query response times. The indexing process is highly efficient, capable of processing 20,000 records against a massive dictionary of 1.47 million terms in approximately 18 minutes.

The primary achievement is the enhanced user experience. The interface allows researchers to find a paper and immediately access associated protein sequences, homologs (via BLAST), and evolutionary trees through embedded links.

Figure 6: The system interface showing search results with mined protein terms linked directly to external protein databases.

Discussion and Future Directions

While the system effectively integrates disparate data sources, challenges remain regarding the precision of text mining due to non-standardized scientific terminology. Future work will focus on developing a formal Chinese viral protein ontology and expanding the system's reach to other gene-specific databases. Additionally, the team plans to implement more advanced machine learning models to further refine mining accuracy and reduce false positives.

Summary updated

BibTeX
@inproceedings{chen2017construction,
  author = {Heng Chen and Yongjuan Zhang and Chunhong Lin and Liwen Zhang and Tao Chen},
  title = {{Construction of Viral Hepatitis Bilingual Bibliographic Database with Protein Text Mining and Information Integration Functions}},
  booktitle = {SML @ IJCAI},
  year = {2017},
}
Figure