Identifying Functions of Proteins in Mice With Functional Embedding Features

Area of Science:

• Computational Biology
• Bioinformatics
• Protein Function Prediction

Background:

• Determining protein functions is crucial in modern biology.
• Experimental methods are insufficient for large-scale protein analysis.
• Computational approaches offer an efficient alternative for function identification.

Purpose of the Study:

• To develop novel computational methods for predicting protein biological functions.
• To introduce new protein representation features using embedding techniques.
• To improve the accuracy and reliability of protein function prediction.

Main Methods:

• Utilized word embedding and network embedding to derive novel features from protein functional domains and protein-protein interaction (PPI) networks.
• Employed the Minimum Redundancy Maximum Relevance (mRMR) method for feature selection.
• Constructed multi-label classifiers using Incremental Feature Selection (IFS) and RAndom k-labELsets (RAKEL).

Main Results:

• Developed two optimal multi-label classifiers based on accuracy and exact match metrics.
• The proposed embedding features significantly enhanced classifier performance.
• Achieved superior results compared to traditional feature extraction methods in protein function prediction.

Conclusions:

• Novel embedding features derived from functional domains and PPI networks are effective for protein function prediction.
• The developed multi-label classification framework provides a robust and accurate approach.
• This computational strategy offers a scalable solution for identifying biological functions of proteins.