Rule generation for protein secondary structure prediction with support vector machines and decision tree

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Support vector machines (SVMs) offer strong generalization for protein structure prediction but lack interpretability.
• Comprehensibility is crucial for machine learning adoption in bioinformatics, aiding experiment guidance and AI integration.
• Decision trees provide good interpretability, a key advantage for complex biological predictions.

Purpose of the Study:

• To develop a novel algorithm, SVM_DT, that combines the strengths of SVMs and decision trees for protein secondary structure prediction.
• To enhance the comprehensibility of SVM models in bioinformatics without sacrificing predictive performance.
• To generate rule sets from SVM_DT for guiding biological experiments and integrating symbolic AI.

Main Methods:

• A three-step algorithm, SVM_DT, was developed.
• Step 1: Train a support vector machine (SVM).
• Step 2: Generate a new training set from the SVM's output.
• Step 3: Train a decision tree on the new set to extract rule sets.

Main Results:

• SVM_DT demonstrated significantly improved comprehensibility compared to standard SVMs.
• The generalization ability of SVM_DT was found to be superior to C4.5 decision trees.
• SVM_DT achieved a generalization performance comparable to that of SVMs.

Conclusions:

• SVM_DT offers a powerful approach for protein secondary structure prediction, balancing high accuracy with enhanced interpretability.
• The extracted rules from SVM_DT can effectively guide "wet experiments" in biological research.
• This hybrid approach facilitates the integration of computational intelligence with symbolic AI for advanced biological data analysis.