Predicting structural class for protein sequences of 40% identity based on features of primary and secondary...

Area of Science:

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

Background:

• Tertiary protein structure discovery lags behind primary structure discovery.
• Machine learning offers a solution to predict protein structural classes and bridge this gap.
• The Structural Classification of Protein - Extended (SCOPe 2.07) dataset is utilized.

Purpose of the Study:

• To predict protein structural classes (α, β, α/β, α + β) using machine learning.
• To leverage primary and secondary protein structure features, including species information.
• To improve the accuracy of protein structural class prediction.

Main Methods:

• Feature extraction from primary and secondary protein structures.
• Inclusion of species information as a predictive feature, validated with UniRef100.
• Training a Random Forest classifier with a 65-dimensional feature vector on a SCOPe 2.07 subset.

Main Results:

• Species information significantly enhances prediction accuracy by 7%-10%.
• Achieved consistent prediction accuracy exceeding 95% on the test set.
• Attained high accuracies on benchmark datasets: ASTRAL 1.73 (>86%), 25PDB (>91%), and FC699 (>97%).

Conclusions:

• Machine learning, incorporating species data, effectively predicts protein structural classes.
• The developed method achieves state-of-the-art accuracy on multiple benchmark datasets.
• This approach contributes to accelerating tertiary structure discovery by improving classification accuracy.