Protein-Protein Interaction Network Extraction Using Text Mining Methods Adds Insight into Autism Spectrum Disorder

Area of Science:

• Biomedical Informatics
• Computational Biology
• Text Mining

Background:

• The rapid expansion of biomedical literature necessitates advanced text mining methods.
• Understanding protein-protein interactions (PPIs) is crucial for elucidating disease mechanisms.
• Existing methods may not precisely distinguish disease-related protein mentions from actual PPIs.

Purpose of the Study:

• To develop an automated system for accurate extraction of PPIs from biomedical texts.
• To improve the precision of identifying sentences specifically describing PPIs.
• To build comprehensive PPI networks for disease-associated proteins.

Main Methods:

• Utilized deep learning, including a BiLSTM recurrent neural network and a conditional random field (CRF) named entity recognition (NER) model.
• Employed a pretrained word embedding and the shortest-dependency path (SDP) model via the SpaCy library.
• Developed a sentence classification model to specifically target PPI-related sentences.

Main Results:

• Achieved a 13% increase in precision for PPI extraction compared to previous BiLSTM models on the Aimed/BioInfr corpus.
• The developed NER model demonstrated 98% precision on the Aimed/BioInfr corpus.
• Successfully mapped protein interactions to complete a curated PPI network for seven Autism Spectrum Disorder-associated proteins.

Conclusions:

• The automated system accurately extracts PPIs, aiding in the comprehension of disease etiology.
• PPI networks generated by this system can illuminate the role of interaction deficits in complex diseases.
• This approach enhances the understanding of how protein interactions influence biological processes and disease development.