A scalable equivariant graph network framework for precise protein function prediction.

Area of Science:

• Computational biology
• Structural bioinformatics
• Machine learning in life sciences

Background:

• Understanding complex cellular processes relies on protein function research.
• Rapid growth in protein sequence data necessitates efficient computational methods for annotation.
• Protein structure and function complexity pose challenges for accurate prediction.

Purpose of the Study:

• To introduce ENGINE, a novel multi-channel deep learning framework for protein function prediction.
• To improve the accuracy and robustness of computational protein annotation.
• To provide biological interpretability for protein function predictions.

Main Methods:

• Utilizing an equivariant graph convolutional network for protein 3D structure geometric features.
• Employing the ESM-C large language model for evolutionary and sequence information.
• Integrating an innovative 3D sequence representation combining spatial and sequential signals.

Main Results:

• ENGINE outperforms state-of-the-art methods in diverse protein function prediction benchmarks.
• The framework demonstrates robust generalization and high predictive accuracy.
• ENGINE provides interpretable insights into critical sequence features and structural motifs, identifying functionally important residues.

Conclusions:

• ENGINE enhances research into cellular processes and disease mechanisms through reliable, interpretable predictions.
• The model facilitates a deeper mechanistic understanding of protein function.
• ENGINE is publicly available as a tool for the scientific community.