Deep_CNN_LSTM_GO: Protein function prediction from amino-acid sequences.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Determining protein functions from amino acid sequences is crucial but resource-intensive.
• Deep learning methods show promise for predicting protein functions computationally.
• Accurate protein function prediction aids in understanding biological systems and disease mechanisms.

Purpose of the Study:

• To propose a novel hybrid deep neural network model, Deep_CNN_LSTM_GO, for predicting protein functions from amino acid sequences.
• To leverage Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) for feature extraction from protein sequences.
• To predict protein functions across the three Gene Ontology (GO) sub-ontologies: Molecular Functions (MF), Biological Process (BP), and Cellular Component (CC).

Main Methods:

• Developed a hybrid deep neural network architecture integrating CNN and LSTM layers.
• Trained and validated the Deep_CNN_LSTM_GO model using the UniProt-SwissProt dataset.
• Evaluated model performance on the Computational Assessment of Function Annotation (CAFA) dataset across MF, BP, and CC sub-ontologies.

Main Results:

• The Deep_CNN_LSTM_GO model demonstrated superior performance compared to existing methods.
• Achieved high accuracy in predicting protein functions across all three Gene Ontology sub-ontologies.
• Outperformed other methods using evaluation metrics Fmax, Smin, and AUPR.

Conclusions:

• The Deep_CNN_LSTM_GO model offers an effective and efficient approach for protein function prediction.
• This hybrid deep learning model significantly advances computational methods for protein functional annotation.
• The findings have implications for accelerating biological research and drug discovery through improved functional genomics.