I-TASSER

I-TASSER predicts three-dimensional protein structures and infers protein functions from amino acid sequences using template-based threading, iterative fragment-assembly simulations, and deep-learning-derived interresidue contact maps.

Key Features:

• Template recognition and structure modeling: Identifies structural templates from the Protein Data Bank (PDB) using multiple threading alignment approaches and generates full-length 3D atomic models via iterative fragment-assembly simulations.
• Integration of evolutionary information: Incorporates interresidue contact maps derived from deep neural-network learning into fragment assembly to improve modeling of proteins without homologous templates, including rapidly mutating viral proteins such as SARS-CoV-2.
• Function prediction: Infers ligand-binding sites, Enzyme Commission (EC) numbers, and Gene Ontology (GO) terms by matching predicted 3D models with known proteins.
• Accuracy estimation: Outputs secondary and tertiary structure predictions with confidence scores that estimate model reliability.
• Atomic-level refinement and local quality estimation: Applies atomic-level structure refinement and local structure quality estimation to enhance high-resolution predictions.

Scientific Applications:

• Structure prediction for novel proteins: Models proteins lacking homologous templates to support studies of novel or uncharacterized sequences.
• Functional annotation: Provides structural-based annotations including ligand-binding sites, EC numbers, and GO terms for predicted proteins.
• Viral protein modeling: Applied to modeling proteins from rapidly mutating viruses such as SARS-CoV-2.
• Structural biology and drug discovery: Supports investigations of protein function and structure-based drug discovery efforts.

Methodology:

Identifies templates from the PDB using multiple threading alignment approaches, assembles full-length 3D atomic models via iterative fragment-assembly simulations, integrates deep neural-network-derived interresidue contact maps with fragment assembly, matches predicted 3D models to known proteins for function inference, and provides secondary/tertiary structures with confidence scores alongside atomic-level refinement and local quality estimation.