Dinosolve: a protein disulfide bonding prediction server using context-based features to enhance prediction accuracy

Area of Science:

• Computational biology
• Protein structure prediction

Background:

• Disulfide bonds are crucial for protein folding and structural stability.
• Accurate prediction of disulfide bonds is vital for understanding protein structure and function.

Purpose of the Study:

• To enhance the accuracy of disulfide bond prediction using context-based features.
• To improve predictions of both disulfide bonding state and connectivity.

Main Methods:

• Derived first-order and second-order mean-force potentials based on amino acid environments around cysteine residues.
• Integrated context-based scores with sequence and evolutionary information.
• Trained neural networks for disulfide bonding state and connectivity prediction.

Main Results:

• Achieved 90.8% accuracy at the residue-level and 85.6% at the protein-level for classifying cysteine residues.
• Demonstrated approximately 2% accuracy improvement in residue-level classification.
• Improved average accuracy for disulfide bonding connectivity prediction with 73.42% sensitivity and 91.61% specificity.

Conclusions:

• Context-based scores effectively enhance disulfide bonding state and connectivity prediction accuracy.
• The developed disulfide prediction algorithm, Dinosolve, is available as a web server.