Protein model accuracy estimation based on local structure quality assessment using 3D convolutional neural network.

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Artificial Intelligence

Background:

• Model quality assessment programs (MQAPs) are crucial for selecting accurate protein tertiary structures from predicted models.
• Traditional 3D convolutional neural networks (3DCNNs) show limitations in protein MQA due to issues like orientation alignment.
• Existing methods struggle with accurate quality evaluation of complex protein structures.

Purpose of the Study:

• To develop a novel single-model MQA method for improved protein tertiary structure prediction.
• To address limitations of existing 3DCNN-based MQA methods by focusing on local structure quality.
• To enhance the accuracy and reliability of selecting the best protein structural models.

Main Methods:

• Proposed a deep neural network incorporating 3DCNN layers for MQA.
• Implemented a local structure quality evaluation for each residue.
• Integrated local quality estimates to assess whole protein structure quality.

Main Results:

• The novel method significantly improved performance across multiple evaluation metrics compared to previous 3DCNN approaches.
• Achieved comparable accuracy to state-of-the-art single-model MQA methods.
• Demonstrated superior performance in CASP11 stage2 with a Pearson coefficient of 0.486, outperforming existing best single-model methods.

Conclusions:

• The proposed local structure quality evaluation method using 3DCNNs offers a significant advancement in protein MQA.
• This approach provides a more accurate and reliable way to select high-quality protein tertiary structure models.
• The method shows promise for improving protein structure prediction pipelines and advancing structural bioinformatics research.