A comprehensive assessment of sequence-based and template-based methods for protein contact prediction

Area of Science:

• Computational biology
• Bioinformatics
• Structural biology

Background:

• Protein contact prediction is crucial for structure modeling.
• Template-based and sequence-based methods exist, but sequence-based methods are often less accurate.
• Assessing contact prediction methods under varying template conditions is necessary.

Purpose of the Study:

• To compare the accuracy of sequence-based (SVM-SEQ) and template-based (SVM-LOMETS) protein contact prediction methods.
• To evaluate the performance of these methods across different target difficulty levels.
• To determine the benefit of combining both prediction approaches.

Main Methods:

• Developed SVM-SEQ, a sequence-based machine learning contact prediction method.
• Developed SVM-LOMETS, a consensus contact prediction method using threading templates.
• Tested both methods on 554 proteins categorized by template similarity (Easy, Medium, Hard, Very Hard).

Main Results:

• SVM-LOMETS outperformed SVM-SEQ on Easy and Medium targets.
• SVM-SEQ showed higher accuracy (12-25%) than SVM-LOMETS on Hard and Very Hard targets.
• Combining SVM-SEQ and SVM-LOMETS increased correct contact predictions by over 60% for Hard targets.

Conclusions:

• State-of-the-art sequence-based contact prediction is valuable for modeling proteins lacking close structural templates.
• Combining sequence- and template-based predictions maximizes accuracy in protein contact prediction.
• This integrated approach can significantly aid tertiary structure modeling efforts.