SDN2GO: An Integrated Deep Learning Model for Protein Function Prediction.

Area of Science:

• Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• Large-scale protein function assignment is vital for biological research.
• Currently, only a small fraction of known proteins have experimentally validated Gene Ontology (GO) annotations.
• This highlights the need for accurate computational methods to predict protein function.

Purpose of the Study:

• To develop an integrated deep-learning model, SDN2GO, for accurate large-scale protein function prediction.
• To leverage diverse biological data including protein sequences, domains, and protein-protein interaction (PPI) networks.
• To improve the prediction accuracy of Gene Ontology (GO) terms.

Main Methods:

• Developed SDN2GO, a deep learning model employing convolutional neural networks (CNNs) for feature extraction from sequences and domains.
• Integrated protein domain information processed using Natural Language Processing (NLP) techniques.
• Utilized a weighted classifier to combine features from sequences, domains, and PPI networks.
• Constructed training and test datasets adhering to the Critical Assessment of Function Annotation (CAFA) time-delayed principle.

Main Results:

• SDN2GO demonstrated superior performance in predicting GO terms across all sub-ontologies compared to two competitive methods and BLAST.
• Incorporating domain features, pre-trained via a deep learning sub-model, significantly enhanced prediction accuracy.
• The model effectively learned and integrated diverse feature types for robust function prediction.

Conclusions:

• SDN2GO offers a highly accurate and effective approach for large-scale protein function prediction.
• The integration of deep learning-derived domain features is a key factor in the model's success.
• The SDN2GO model and associated scripts are available as open-source software, facilitating further research.