A novel hybrid CNN and BiGRU-Attention based deep learning model for protein function prediction

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning in Biology

Background:

• Understanding protein function is crucial for deciphering life's molecular mechanisms.
• Deep learning models like CNN, GRU, and LSTM offer distinct advantages in sequence analysis.
• Protein Language Models (PLMs) leverage attention networks for effective protein sequence representation.

Purpose of the Study:

• To develop a hybrid deep-learning model integrating CNN and BiGRU-Attention with PLM embeddings for enhanced protein function prediction.
• To combine the strengths of CNNs (short-term dependencies) and BiGRU-Attention (long-term dependencies) for comprehensive sequence analysis.

Main Methods:

• Proposed a hybrid model combining Convolutional Neural Networks (CNN) and Bidirectional Gated Recurrent Unit (BiGRU)-Attention.
• Incorporated protein language model embeddings to enrich sequence representations.
• Evaluated the model's performance on human and yeast datasets for protein function prediction tasks.

Main Results:

• The hybrid CNN + BiGRU-Attention model with PLM embeddings demonstrated improved Fmax scores compared to the state-of-the-art SDN2GO model.
• Significant performance gains were observed across cellular component, molecular function, and biological process prediction tasks for both human and yeast datasets.
• Specific improvements for the human dataset were 1.9% (cellular component), 3.8% (molecular function), and 0.6% (biological process).
• Specific improvements for the yeast dataset were 2.4% (cellular component), 5.2% (molecular function), and 1.2% (biological process).

Conclusions:

• The proposed hybrid deep-learning approach effectively integrates diverse sequence modeling capabilities for accurate protein function prediction.
• The model's superior performance highlights the benefit of combining CNN, BiGRU-Attention, and PLM embeddings.
• This work contributes a more effective computational tool for advancing the understanding of protein functions.