Protein tertiary structure modeling driven by deep learning and contact distance prediction in CASP13

Area of Science:

• Computational Biology
• Structural Biology
• Bioinformatics

Background:

• Protein structure prediction is crucial for understanding biological function.
• Deep learning has revolutionized computational biology since 2012.
• Accurate prediction of residue-residue contacts is a key challenge.

Purpose of the Study:

• To enhance the MULTICOM protein structure prediction system for CASP13.
• To investigate the impact of deep learning on contact distance prediction and model ranking.
• To address challenges in protein structure prediction.

Main Methods:

• Utilized deep convolutional neural networks for contact distance prediction.
• Employed distance-driven template-free (ab initio) modeling.
• Integrated deep learning and contact prediction for protein model ranking.

Main Results:

• MULTICOM ranked 3rd out of 98 predictors in CASP13 (template-free and template-based).
• Deep convolutional neural networks improved contact distance prediction using global features.
• Deep learning enhanced model quality assessment and ranking.

Conclusions:

• Deep learning and contact distance prediction are key to solving protein structure prediction.
• Contact prediction consistently improved model ranking.
• Challenges remain in predicting contacts with limited sequence data and noisy predictions.