Prediction of turn types in protein structure by machine-learning classifiers.

Area of Science:

• Computational biology
• Structural bioinformatics
• Machine learning in protein science

Background:

• Protein secondary structures, specifically turns, are crucial for protein folding and function.
• Accurate prediction of turns from amino acid sequences remains a challenge in bioinformatics.

Purpose of the Study:

• To develop and evaluate machine learning approaches for predicting protein turns and their types directly from amino acid sequences.
• To introduce and validate a novel classification scheme for protein turns.

Main Methods:

• Training an unsupervised self-organizing map and two kernel-based classifiers: support vector machine (SVM) and probabilistic neural network (PNN).
• Classifying turn versus non-turn sequences within families characterized by intramolecular hydrogen bonds and lengths of three to six residues.
• Utilizing PNNs for predicting specific beta-turn types.

Main Results:

• Support vector machine classifiers achieved approximately 80% prediction accuracy and a Matthews correlation coefficient (MCC) of 0.6 for turn prediction.
• Probabilistic neural networks successfully distinguished between five types of beta-turns, yielding MCC > 0.5 and over 80% overall accuracy.

Conclusions:

• The proposed novel turn classification scheme is well-defined and effective.
• Machine learning classifiers, particularly SVM and PNN, are well-suited for sequence-based prediction of protein turns and beta-turn types.
• These methods hold significant potential for predicting protein structures from sequences.