Using word embedding technique to efficiently represent protein sequences for identifying substrate specificities of...

Area of Science:

• Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• Membrane transport proteins are vital for cellular functions.
• Understanding their substrate specificities is crucial for drug design and disease analysis.

Purpose of the Study:

• To apply natural language processing (NLP) techniques to predict transporter substrate specificities.
• To develop novel feature representation methods for protein sequences.

Main Methods:

• Utilized word embedding and frequency analysis on biological words within protein sequences.
• Optimized feature extraction by varying biological word lengths.
• Employed machine learning models for prediction tasks.

Main Results:

• Achieved superior performance compared to existing feature types.
• Reached an average area under the curve (AUC) of 0.96 (5-fold cross-validation) and 0.99 (independent test).

Conclusions:

• The proposed NLP-based feature extraction method significantly improves substrate specificity prediction for transporters.
• This work provides a foundation for applying NLP in bioinformatics and aids in transporter identification.