BERT-DomainAFP: Antifreeze protein recognition and classification model based on BERT and structural domain...

Area of Science:

• Biochemistry and Molecular Biology
• Bioinformatics and Computational Biology

Background:

• Antifreeze proteins (AFPs) are vital for organisms' adaptation to cold environments.
• AFPs have broad applications in medicine, food science, aquaculture, and agriculture.
• Accurate identification of AFPs is hindered by their structural and sequence diversity.

Purpose of the Study:

• To develop an advanced deep learning model for improved antifreeze protein (AFP) prediction and classification.
• To enhance the accuracy and efficiency of AFP recognition for research and industrial applications.

Main Methods:

• Development of the BERT-DomainAFP model, a deep learning approach utilizing pre-trained ProteinBERT.
• Creation of the AntiFreezeDomains dataset with a novel annotation strategy.
• Implementation of oversampling and undersampling techniques to address data imbalance.

Main Results:

• The BERT-DomainAFP model achieved a 98.48% accuracy rate, surpassing existing methods.
• The model demonstrated the ability to classify different AFP types based on structural domain features.
• The developed model offers superior performance in AFP recognition and classification.

Conclusions:

• BERT-DomainAFP provides a highly accurate and effective solution for antifreeze protein identification and classification.
• The model's performance indicates a significant advancement in computational approaches for studying AFPs.
• This tool holds promise for accelerating research and applications involving antifreeze proteins.