A deep-learning system bridging molecule structure and biomedical text with comprehension comparable to human...

Area of Science:

• Biomedical Informatics
• Artificial Intelligence in Medicine
• Computational Chemistry

Background:

• Biomedical research relies on understanding complex molecule entities.
• Current deep learning models process molecule structures and biomedical text data separately, limiting comprehensive knowledge acquisition.
• Accelerating biomedical research necessitates advanced machine reading systems capable of integrating diverse data types.

Purpose of the Study:

• To develop a unified deep-learning framework for machine reading that integrates both molecule structure and biomedical text information.
• To overcome the limitations of existing models that process data types in isolation.
• To provide comprehensive biomedical research assistance and facilitate applications like molecular property prediction and relation extraction.

Main Methods:

• A novel knowledgeable machine reading system was developed using a unified deep-learning framework.
• The system employs unsupervised learning to grasp meta-knowledge within and across different information sources (molecule structures and biomedical text).
• The framework enables simultaneous processing of diverse data types for a holistic understanding of molecule entities.

Main Results:

• The developed system achieves a comprehensive understanding of molecule entities by bridging structural and textual data.
• Experimental results demonstrate that the system surpasses human professionals in molecular property comprehension.
• The system shows significant potential for facilitating automatic drug discovery and documentation.

Conclusions:

• The unified deep-learning framework effectively integrates diverse biomedical data for enhanced molecule understanding.
• This approach significantly advances machine reading capabilities in biomedical research.
• The system holds promising potential for accelerating drug discovery and improving biomedical documentation processes.