DeepGOPlus: improved protein function prediction from sequence.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Protein function prediction is vital for understanding biological processes and identifying drug targets.
• Existing methods often rely on features beyond sequence, which are difficult to obtain or unavailable for many proteins.
• Sequence-based methods can be less accurate and time-consuming compared to multi-feature approaches.

Purpose of the Study:

• To develop a novel, accurate, and fast method for protein function prediction using only protein sequence data.
• To overcome limitations of existing methods by integrating deep learning with sequence similarity.
• To provide a scalable solution for annotating a large number of protein sequences.

Main Methods:

• Developed DeepGOPlus, a novel method combining a deep convolutional neural network (CNN) with sequence similarity predictions.
• The CNN model identifies function-predictive motifs within protein sequences.
• Integrated motif information with functions of homologous proteins to enhance prediction accuracy.

Main Results:

• DeepGOPlus achieved high performance on CAFA3 evaluations, with Fmax scores of 0.390 (BPO), 0.557 (MFO), and 0.614 (CCO).
• These scores position DeepGOPlus among the top-performing methods, particularly excelling in Biological Process (BPO) and Molecular Function (MFO) predictions.
• The method demonstrates high efficiency, annotating approximately 40 protein sequences per second on standard hardware.

Conclusions:

• DeepGOPlus offers a powerful and efficient solution for protein function prediction solely from sequence data.
• The integration of CNNs and sequence similarity effectively addresses limitations of traditional methods.
• This advancement facilitates large-scale, accurate protein function annotation for diverse biological applications.