The I-TASSER Suite: protein structure and function prediction

The I-TASSER Suite is a stand-alone package for predicting protein structure and function. It identifies low-free-energy conformations by structure clustering, runs a second assembly simulation from centroid models to remove steric clashes, and builds final atomic models through two-step energy minimization. Model reliability is estimated with a confidence score based on threading alignments and clustering density, while the newly developed ResQ method reports residue-level local quality and B-factors. For function, top models are matched against the BioLiP database, and three complementary algorithms (COFACTOR, TM-SITE, S-SITE) are combined by COACH to infer ligand-binding sites, Enzyme Commission numbers, and Gene Ontology terms.

The pipeline was validated in community-wide blind experiments including CASP10 and CAMEO. I-TASSER produced correct folds (TM-score above 0.5) for 10 of 36 'New Fold' targets lacking homologous templates and for 92 of 110 template-based targets, while pulling templates closer to native structures. In CAMEO, the COACH function predictor generated ligand-binding-site predictions for 4,271 targets at an average accuracy of 0.86, about 20% higher than the next-best method. By packaging accurate structure modeling and function annotation together, the Suite became a widely used off-line tool for structural bioinformatics.