Ab initio modeling of small proteins by iterative TASSER simulations

Area of Science:

• Computational structural biology
• Protein structure prediction
• Bioinformatics

Background:

• Predicting 3D protein structures from amino acid sequences is a major challenge in computational structural biology.
• Ab initio (from scratch) modeling is particularly difficult for proteins lacking homologous structures in the Protein Data Bank (PDB).
• Existing ab initio methods have shown slow progress.

Purpose of the Study:

• To extend the TASSER (threading/assembly/refinement) method for ab initio modeling.
• To systematically evaluate the ability of the extended method, I-TASSER, to fold small single-domain proteins.

Main Methods:

• Iterative implementation of the TASSER method to develop I-TASSER.
• Testing I-TASSER on three benchmarks of small proteins with varying sizes and homology.

Main Results:

• I-TASSER models for 16 small proteins (<90 residues) achieved an average Calpha-RMSD of 3.8A, with 6 below 2.5A.
• For 20 small proteins (<120 residues), I-TASSER yielded an average Calpha-RMSD of 3.9A, outperforming TOUCHSTONE-II (5.9A).
• On 20 non-homologous PDB proteins (<120 residues), I-TASSER achieved an average Calpha-RMSD of 3.9A, with 7 cases below 2.5A.

Conclusions:

• I-TASSER consistently predicts correct folds and high-resolution models for small single-domain proteins.
• I-TASSER demonstrates superior or comparable performance to ROSETTA and TOUCHSTONE II with significantly lower computational cost.
• The success of the automated I-TASSER server in CASP7 highlights advancements in automated ab initio model generation.