A retrospective analysis of CASP2 threading predictions

Area of Science:

• Computational biology
• Structural bioinformatics
• Protein structure prediction

Background:

• Protein threading methods are crucial for predicting protein structures.
• The accuracy of protein structure prediction varies significantly across different targets.
• Assessing the performance of threading methods requires understanding target difficulty.

Purpose of the Study:

• To define critical thresholds for fold recognition specificity and threading model accuracy in CASP2.
• To identify key factors determining the success rate of protein structure predictions.
• To propose a method for characterizing the sustained performance of threading methods based on target difficulty.

Main Methods:

• Analysis of CASP2 protein threading results.
• Setting critical thresholds for fold recognition specificity and threading model accuracy.
• Quantifying the percentage of superimposable residues and residue identity in structural alignments.
• Categorizing targets based on increasing difficulty.

Main Results:

• Correct structure predictions were achieved for targets with over 50% superimposable residues.
• Three-fourths of correct predictions had >12% residue identity and sequence motifs.
• A strong correlation was observed between target difficulty and prediction success.
• Established thresholds helped differentiate correct from incorrect predictions.

Conclusions:

• Protein structure prediction success is highly dependent on target characteristics.
• Target difficulty, defined by residue superposition and identity, is a critical factor for threading method performance.
• Characterizing threading method performance by counting correct predictions for increasing target difficulty provides a robust assessment.