A CNN-CBAM-BIGRU model for protein function prediction.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Predicting protein function from amino acid sequences is a fundamental challenge in bioinformatics.
• Deep learning models have shown promise in automating feature extraction for protein function prediction.
• Existing methods often struggle to capture complex sequence dependencies.

Purpose of the Study:

• To propose a novel deep learning model, CNN-CBAM+BiGRU, for improved protein function prediction.
• To leverage attention mechanisms for more effective feature extraction from protein sequences.
• To enhance the capture of long-range dependencies within protein sequences.

Main Methods:

• The study introduces a hybrid model combining Convolutional Neural Networks (CNNs) with a Convolutional Block Attention Module (CBAM) and Bidirectional Gated Recurrent Units (BiGRUs).
• CBAM guides the CNN to focus on salient regions of protein sequences, improving feature representation.
• BiGRUs are employed to effectively model long-range sequential dependencies crucial for function prediction.

Main Results:

• The CNN-CBAM+BiGRU model demonstrated superior performance compared to the CNN-BIGRU+ATT model on both human and yeast datasets.
• For the human dataset, improvements of +1.0% (cellular components), +1.1% (molecular functions), and +0.5% (biological processes) were observed.
• For the yeast dataset, the model achieved higher accuracy with gains of +2.4% (cellular component), +1.2% (molecular functions), and +0.6% (biological processes).

Conclusions:

• The integration of CBAM and BiGRU significantly enhances the accuracy of protein function prediction.
• The proposed model offers a more effective approach to feature extraction and dependency modeling in protein sequences.
• This advancement holds potential for accelerating biological research through more precise functional annotation of proteins.