TopModel

TopModel predicts protein three-dimensional structures and refines template-based models using a top-down consensus approach combined with deep neural networks to bridge the gap between next-generation sequencing-derived protein sequences and experimentally determined structures.

Key Features:

• Top-down consensus with deep neural networks: Combines a top-down consensus approach with deep neural networks for improved template selection and error correction.
• Template selection strategy: Selects templates more effectively than majority-vote and model-averaging strategies, reducing deviations from the native fold.
• Mis-modeled region correction: Identifies and corrects incorrectly modeled regions in predicted structures.
• Threading, alignment, and model quality estimation: Integrates threading, alignment, and model quality estimation techniques within a single workflow.
• Template-based prediction workflow: Provides a versatile workflow and toolbox tailored for template-based protein structure prediction.
• CASP benchmarking: Demonstrated superior performance in template selection, alignment accuracy, and overall model quality on CASP10-12 datasets compared to 12 leading primary predictors.
• Prospective validation: Produced prospective predictions of the nisin resistance protein (NSR) from Streptococcus agalactiae and LipoP from Clostridium difficile that agreed better with experimental data than constituent primary predictors.
• Integration with experimental data: Can be combined with sparse or low-resolution experimental data to refine final models.

Scientific Applications:

• Protein function annotation: Improves structural models used to infer protein function.
• Evolution and stability studies: Supports analyses of protein evolution, dynamics, and stability through improved structural models.
• Interaction analysis: Aids study of protein–protein interactions and interaction interfaces by providing higher-fidelity models.
• Data-driven design: Enables data-driven protein and drug design by improving model accuracy for design applications.
• Integrative structural biology: Refines models by integrating sparse or low-resolution experimental data.
• Method benchmarking: Serves in benchmarking and method development as demonstrated on CASP10-12 datasets.
• Prospective bacterial protein prediction: Applied to prospective structure prediction of bacterial proteins such as NSR and LipoP to compare with experimental data.

Methodology:

Combines a top-down consensus approach with deep neural networks for template selection and correction, and integrates threading, alignment, and model quality estimation for template-based protein structure prediction.