A Deep Learning Framework for Gene Ontology Annotations With Sequence- and Network-Based Information.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Accurate protein function annotation is crucial for biology and medicine.
• High-throughput technologies have identified numerous proteins, but many lack functional annotations.
• Proteins often possess multiple functions and participate in complex interactions, necessitating advanced prediction methods.

Purpose of the Study:

• To develop a deep learning framework, DeepGOA, for predicting protein functions.
• To integrate protein sequence information and protein-protein interaction (PPI) networks for enhanced prediction accuracy.
• To address the challenge of annotating proteins with unknown functions.

Main Methods:

• Utilized word2vec, Bi-directional Long and Short Time Memory (Bi-LSTM), and multi-scale Convolutional Neural Networks (CNN) for protein sequence feature extraction.
• Employed the InterPro tool to identify subsequence-based features like domains and motifs.
• Applied the Deepwalk algorithm for generating embeddings from PPI networks.
• Concatenated sequence-derived and PPI-derived features for function prediction.

Main Results:

• DeepGOA demonstrated superior performance in predicting protein functions compared to existing methods like DeepGO and BLAST.
• The framework effectively integrates diverse data sources (sequences and PPIs) for robust functional annotation.
• Experimental evaluations confirmed the efficacy of the proposed deep learning architecture.

Conclusions:

• DeepGOA offers a powerful and accurate approach for protein function prediction.
• The integration of sequence semantics, subsequence features, and PPI network information significantly improves annotation.
• This framework has the potential to accelerate biological discovery and advance medical research by elucidating protein roles.