PEDL: extracting protein-protein associations using deep language models and distant supervision

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• Molecular biology research relies on comprehensive protein-protein association (PPA) data for understanding signaling pathways.
• Existing pathway databases are often incomplete, hindering research progress.
• Current PPA extraction methods are limited by their reliance on scarce manually labeled data.

Purpose of the Study:

• To develop an automated method for extracting functional protein-protein associations (PPAs) from biomedical literature.
• To improve the completeness and accuracy of pathway databases using advanced natural language processing techniques.

Main Methods:

• Proposed PPA Extraction with Deep Language (PEDL), a novel method combining deep language models and distant supervision.
• Utilized three newly introduced datasets for PPA prediction and text span identification.
• Leveraged distant supervision to access a significantly larger training dataset compared to manual annotation-dependent methods.

Main Results:

• PEDL demonstrated superior performance over a state-of-the-art model in both PPA prediction and text span identification tasks across all evaluated datasets.
• Expert evaluation confirmed PEDL's capability to identify PPAs absent in major pathway databases.
• The method accurately pinpointed the specific text spans supporting the predicted PPAs.

Conclusions:

• PEDL offers a scalable and effective solution for enhancing the comprehensiveness of protein-protein association data.
• The developed method can significantly contribute to a more complete understanding of biological signaling pathways.
• PEDL provides a valuable tool for researchers aiming to augment existing pathway databases with novel associations extracted from literature.