Protein structure prediction begins well but ends badly

Area of Science:

• Computational Biology
• Structural Biology
• Bioinformatics

Background:

• Accurate protein structure prediction remains a significant challenge in molecular biology.
• Current prediction algorithms often treat all amino acid residues uniformly, irrespective of their position in the sequence.

Purpose of the Study:

• To investigate positional variations in protein structure prediction accuracy.
• To determine if predictive capability differs between N-terminal and C-terminal regions of protein sequences.

Main Methods:

• Analysis of blind prediction data from Critical Assessment of Techniques for Protein Structure Prediction (CASP) experiments.
• Evaluation of data from EVA, an automated secondary structure prediction server test.
• Comparison of prediction accuracy for N-terminal and C-terminal fragments based on root-mean-square deviation (RMSD).

Main Results:

• Protein structure prediction accuracy is demonstrably dependent on sequence position.
• Both secondary and tertiary structure predictions exhibit higher accuracy in N-terminal regions compared to C-terminal regions.
• Eight out of ten secondary structure prediction algorithms evaluated by EVA showed significantly better performance on N-terminal regions.
• CASP data confirmed this bias, with N-terminal fragments predicted more accurately than C-terminal fragments in 79% of cases at a 20-residue fragment length.

Conclusions:

• Positional bias exists in current protein structure prediction methods.
• N-terminal regions of proteins are generally predicted with higher accuracy than C-terminal regions.
• Future prediction algorithms may need to account for these positional dependencies for improved accuracy.