Predicting residue-wise contact orders in proteins by support vector regression.

Area of Science:

• Protein bioinformatics
• Computational biology
• Structural bioinformatics

Background:

• Residue-wise contact order (RWCO) quantifies long-range contacts in protein sequences.
• RWCO is a generalization of contact order, crucial for 3D structure reconstruction.
• Predicting RWCO aids in protein structure and folding rate prediction, and understanding sequence-structure relationships.

Purpose of the Study:

• To develop a novel computational approach for predicting protein residue-wise contact order (RWCO) values.
• To evaluate the impact of various sequence encoding schemes on RWCO prediction accuracy.
• To establish a robust method for extracting protein sequence-structure relationships.

Main Methods:

• Employed support vector regression (SVR) for predicting RWCO from primary amino acid sequences.
• Investigated seven sequence encoding schemes, including PSI-BLAST profiles, amino acid composition, molecular weight, and predicted secondary structure.
• Utilized a dataset of 680 protein sequences for model training and validation.

Main Results:

• Achieved a Pearson correlation coefficient (CC) of 0.55 and RMSE of 0.82 using PSI-BLAST profiles.
• Improved prediction accuracy to CC 0.57 and RMSE 0.79 by incorporating global features (molecular weight, amino acid composition).
• Reached the best performance with CC 0.60 and RMSE 0.78 by combining predicted secondary structure (PSIPRED), outperforming existing methods.

Conclusions:

• Support vector regression (SVR) provides a powerful and accurate method for predicting RWCO values.
• The SVR approach demonstrates competitive performance compared to linear regression-based methods.
• This study highlights SVR's efficacy in estimating protein structural profiles from sequences.