Integrating ESM‑2 and Graph Neural Networks with AlphaFold‑2 Structures for Enhanced Protein Function Prediction

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Accurate protein function prediction is crucial for biological research and drug discovery.
• Current deep learning methods face limitations, especially for proteins lacking interaction data or when treating sequence and structural information separately.

Purpose of the Study:

• To develop an improved graph-based framework for protein function prediction.
• To integrate advanced sequence embeddings from ESM-2 and enhanced structural features from AlphaFold2-predicted structures.

Main Methods:

• Utilized ESM-2, a state-of-the-art protein language model, for generating rich sequence embeddings.
• Implemented a hybrid pooling mechanism in graph convolutional blocks to capture global and local structural features.
• Applied the framework to the human proteome for function annotation.

Main Results:

• The proposed model consistently outperformed existing methods on the human proteome.
• Demonstrated superior performance in predicting molecular function, cellular component, and biological process annotations.
• Highlighted the benefits of integrating sequence and structural information for enhanced prediction accuracy.

Conclusions:

• The combined approach of advanced sequence representations and enhanced structural learning offers a powerful strategy for accurate protein function prediction.
• This framework addresses limitations of previous methods by effectively integrating diverse data modalities.
• The findings pave the way for more comprehensive understanding of protein functions and accelerate biological discovery.