BBATProt: a framework predicting biological function with enhanced feature extraction via interpretable deep...

Area of Science:

• Computational Biology
• Machine Learning in Bioinformatics
• Protein Function Prediction

Background:

• Experimental methods for protein function determination are limited.
• Computational approaches, especially machine learning, are crucial for predicting protein and peptide functions.
• Existing tools often lack versatility and are task-specific.

Purpose of the Study:

• To develop a versatile framework for predicting protein and peptide functions.
• To improve the accuracy and adaptability of computational function prediction tools.
• To leverage transfer learning and advanced neural network architectures for enhanced prediction.

Main Methods:

• Proposed the BERT-BiLSTM-Attention-TCN Protein Function Prediction Framework (BBATProt).
• Utilized transfer learning with a pretrained bidirectional encoder representations from transformer (BERT) model.
• Integrated bidirectional long short-term memory (BiLSTM) and temporal convolutional network (TCN) with attention mechanisms.

Main Results:

• BBATProt demonstrated superior performance over state-of-the-art models in various prediction tasks.
• Achieved significant accuracy improvements in antimicrobial peptide (AMP) prediction (2.96%-41.96%) and post-translational modification (PTM) site prediction (0.64%-23.54%).
• Showcased improved area under the receiver operating characteristic curve for AMP (0.71%-40.51%) and PTM (0.62%-27.82%) predictions.
• Attention mechanism visualizations confirmed framework interpretability and feature extraction insights.

Conclusions:

• BBATProt offers a versatile and accurate solution for protein and peptide function prediction.
• The framework's architecture effectively captures both local and global features for precise predictions.
• The interpretability of BBATProt provides valuable insights into the underlying biological properties and prediction mechanisms.