Fast and effective protein model refinement using deep graph neural networks

Area of Science:

• Computational biology
• Structural bioinformatics
• Artificial intelligence in biology

Background:

• Protein model refinement is crucial for improving predicted protein structures.
• Current methods, relying on extensive sampling, are computationally expensive and time-consuming.

Purpose of the Study:

• To develop a fast and effective protein model refinement method.
• To leverage graph neural networks (GNNs) for predicting inter-atom distance distributions.

Main Methods:

• Utilized graph neural networks (GNNs) to predict refined inter-atom distance probability distributions.
• Rebuilt 3D protein models from the predicted distance distributions.
• Evaluated performance on Critical Assessment of Structure Prediction (CASP) refinement targets.

Main Results:

• The GNN-based method achieved accuracy comparable to leading human groups (Feig and Baker).
• Refinement time was drastically reduced: ~11 minutes per model on 1 CPU, compared to hours for Baker and Feig.
• GNNs outperformed ResNet (convolutional residual neural networks) in refinement with limited conformational sampling.

Conclusions:

• Graph neural networks offer a significantly faster and effective approach to protein model refinement.
• This method presents a viable alternative to traditional, time-intensive refinement techniques.
• The study highlights the potential of GNNs in accelerating structural bioinformatics workflows.