Identification of Moonlighting Proteins from Published Literature Using Natural Language Processing and AI.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Functional genomics research is expanding, necessitating efficient methods for protein functional annotation and understanding their diverse roles.
• Existing annotation methods often overlook valuable information scattered across scientific literature.
• Protein moonlighting, where proteins perform multiple functions, presents a challenge due to functional redundancy.

Purpose of the Study:

• To develop and validate a natural language processing (NLP) model for identifying protein moonlighting behavior through literature mining.
• To address the limitations of current knowledge-driven and data-driven approaches by leveraging existing scientific text.
• To improve the accuracy of predicting protein functional properties by incorporating literature-derived insights.

Main Methods:

• Utilized a PubMed BERT model, pre-trained on biomedical literature, and further optimized it through targeted retraining.
• Applied the optimized NLP model to mine scientific publications for evidence of moonlighting functions in proteins.
• Compared the performance of the NLP-based approach against established first-principle methods.

Main Results:

• The NLP model demonstrated high accuracy in identifying proteins exhibiting moonlighting behavior.
• The literature mining approach significantly outperformed previous methods for predicting moonlighting functions.
• The model's effectiveness was validated on specific datasets, showcasing its predictive power.

Conclusions:

• NLP-based literature mining offers a powerful and scalable solution for identifying protein moonlighting.
• This approach enhances functional annotation by harnessing dispersed information from scientific publications.
• The developed methods are applicable to broader literature-mining challenges within the biological domain.