Document triage for identifying protein-protein interactions affected by mutations: a neural network ensemble...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Precision medicine (PM) aims to tailor treatments based on individual genetic profiles.
• Identifying protein-protein interactions (PPI) affected by genetic mutations is crucial for PM.
• Biomedical literature mining is essential for extracting this complex information.

Purpose of the Study:

• To develop and evaluate a computational approach for identifying scientific literature relevant to mutations affecting protein-protein interactions (PPI).
• To improve the accuracy of document triage for precision medicine research using machine learning.

Main Methods:

• A neural network ensemble model was developed for document classification.
• Multiple neural network models were combined for improved performance in identifying relevant articles.
• A protein-protein interaction (PPI) pre-trained module was incorporated to enhance the model, especially with limited training data.

Main Results:

• The ensemble approach outperformed individual models in the BioCreative VI PM document triage task.
• The best submission achieved an F-score of 69.04% in the official runs.
• Post-challenge analysis and incorporation of a PPI pre-trained module improved the F-score to 71.04% on the test set.

Conclusions:

• The neural network ensemble method is effective for mining biomedical literature on mutation-affected PPIs.
• The approach aids researchers in the precision medicine endeavor by identifying relevant studies.
• Further improvements were achieved by incorporating pre-trained modules, addressing data limitations.