A Novel Automated Approach to Mutation-Cancer Relation Extraction by Incorporating Heterogeneous Knowledge

Area of Science:

• Bioinformatics
• Computational Biology
• Natural Language Processing

Background:

• Precision cancer medicine relies on understanding gene mutation-cancer relationships.
• Extracting these relationships from scientific literature is challenging due to the need for domain-specific knowledge.
• Existing text mining methods often struggle with the complexity of mutation-cancer associations.

Purpose of the Study:

• To develop a deep learning model for the joint extraction of gene mutations and their associated cancers.
• To enhance mutation-cancer relation extraction by integrating knowledge from diverse biological knowledge bases.
• To improve the accuracy and robustness of text mining for cancer research.

Main Methods:

• Proposed a deep learning model for joint mutation and cancer entity extraction.
• Implemented two novel knowledge integration methods: sentence-based and attribute-aware embedding.
• Utilized two distinct knowledge bases containing mutation-specific information.

Main Results:

• Achieved high F1 scores: 96.00% on EMU BCa, 92.57% on EMU PCa, and 94.57% on BRONCO.
• Demonstrated superior performance compared to several baseline models.
• Showcased the model's ability to leverage knowledge bases for accurate linking, even with limited contextual text.

Conclusions:

• The proposed deep learning model with integrated knowledge bases significantly advances mutation-cancer relation extraction.
• The novel knowledge embedding strategies are effective in capturing complex biological relationships.
• This approach supports precision cancer medicine by providing reliable, extracted information from biomedical literature.