Dimensionality reduction for protein secondary structure and solvent accesibility prediction

Area of Science:

• Bioinformatics
• Computational Biology
• Structural Biology

Background:

• Protein secondary structure and solvent accessibility prediction are crucial for determining 3D protein structures.
• Increasingly sophisticated feature extraction methods lead to high-dimensional input spaces.
• High dimensionality can cause slow training, prediction, and noise, hindering model performance.

Purpose of the Study:

• To investigate the effectiveness of dimensionality reduction techniques for protein structure prediction.
• To reduce the input feature space for secondary structure and solvent accessibility prediction.
• To enable the inclusion of more features without compromising prediction accuracy.

Main Methods:

• Applied various dimensionality reduction techniques: feature selection, autoencoders, and Principal Component Analysis (PCA).
• Utilized the reduced feature set to train a Support Vector Machine (SVM) within a hybrid classifier.
• Evaluated performance using cross-validation on two challenging benchmark datasets.

Main Results:

• Demonstrated substantial reduction in input feature space dimensionality.
• Maintained high prediction accuracy for protein secondary structure and solvent accessibility.
• Confirmed that dimensionality reduction effectively combats overfitting.

Conclusions:

• Dimensionality reduction is a viable strategy for improving protein structure prediction efficiency and accuracy.
• Allows for the integration of novel, informative features without performance degradation.
• Facilitates more robust and accurate prediction of protein structural properties.