PFmulDL: a novel strategy enabling multi-class and multi-label protein function annotation by integrating diverse...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Protein function annotation is crucial but challenging, with existing methods struggling to accurately classify proteins in rare families.
• Current approaches often overemphasize large protein families, leading to misclassification of proteins in underrepresented groups.

Purpose of the Study:

• To develop an advanced protein function annotation strategy that addresses the limitations of existing methods, particularly for rare protein classes.
• To enhance the accuracy and scope of protein function prediction using integrated deep learning techniques.

Main Methods:

• Developed PFmulDL, a novel strategy integrating recurrent neural networks (RNN) and convolutional neural networks (CNN) for protein function annotation.
• Incorporated transfer learning to further boost prediction performance.
• Utilized the latest Gene Ontology data for model training and validation.

Main Results:

• PFmulDL successfully annotates the largest number of protein families compared to existing methods.
• The strategy significantly improves prediction performance for rare protein classes.
• Performance for major protein classes is maintained without degradation.

Conclusions:

• PFmulDL offers a significant advancement in protein function annotation, particularly for underrepresented protein families.
• This strategy serves as a vital complement to existing methods, addressing the need for improved prediction of rare class proteins.
• The developed models and source code are publicly available to facilitate further research and application.