Towards structured output prediction of enzyme function.

Area of Science:

• Bioinformatics
• Computational Biology
• Enzyme Function Prediction

Background:

• Developing kernel methods for predicting enzyme function.
• Focus on structured output prediction for enzymatic reactions.
• Utilizing Hierarchical Max-Margin Markov (HM3) and Maximum Margin Regression (MMR) algorithms.

Purpose of the Study:

• Compare HM3 and MMR for hierarchical enzyme function classification.
• Evaluate the effectiveness of various sequence features, including string kernels and GTG features.
• Assess the accuracy and feasibility of structured output prediction for enzyme function.

Main Methods:

• Employed Hierarchical Max-Margin Markov (HM3) and Maximum Margin Regression (MMR) for structured output prediction.
• Utilized string kernels and Global Alignment Trace Graph (GTG) features derived from protein sequence alignments.
• Performed hierarchical classification of enzyme EC numbers and Gold Standard enzyme families.

Main Results:

• Achieved over 85% accuracy for EC classification and over 79% for enzyme families.
• Obtained over 91% microlabel F1 score for EC digits and over 89% for enzyme families.
• Demonstrated that GTG features with a polynomial kernel are essential for accurate prediction, outperforming nearest neighbor classifiers.

Conclusions:

• Structured output prediction with GTG features is computationally feasible.
• The developed methods achieve accuracy comparable to state-of-the-art approaches in enzyme function prediction.