Predicting dihedral angle probability distributions for protein coil residues from primary sequence using neural...

Area of Science:

• Bioinformatics
• Computational Biology
• Structural Biology

Background:

• Protein structure prediction from amino acid sequence is a major bioinformatics challenge.
• Random coil segments (~40% of proteins) lack clear patterns, hindering prediction accuracy.
• Flanking residues influence coil segment dihedral angles, suggesting non-random structure.

Purpose of the Study:

• To predict the probability distribution of dihedral angles in protein coil segments based on flanking residues.
• To improve the accuracy of protein structure prediction methods.

Main Methods:

• Developed an artificial neural network using an amino acid input window.
• The network predicts a dihedral angle probability distribution for the central residue.
• Trained the network on existing protein data.

Main Results:

• Achieved significant improvement (4-68%) in predicting the most probable dihedral angle bin compared to baseline statistics.
• Attained accuracy comparable to secondary structure prediction (~80%) by considering the top 20 predicted bins.
• Demonstrated the model's ability to capture local context-dependent dihedral angle propensities.

Conclusions:

• The predicted dihedral angle distributions can enhance tertiary structure prediction methods that sample backbone angles.
• This approach offers potential improvements for fragment assembly methods in protein structure prediction.
• Provides a novel way to predict local structure context for previously unstructured protein regions.