POIMs: positional oligomer importance matrices--understanding support vector machine-based signal detectors.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Biological sequence classification is crucial for gene finding and function prediction.
• Support Vector Machines (SVMs) with complex sequence kernels are accurate but lack interpretability.
• Understanding SVM decision rules is challenging for biological interpretation.

Purpose of the Study:

• To enhance the accessibility and utility of SVM-based sequence classifiers.
• To develop a method for visualizing and interpreting sequence patterns identified by SVMs.
• To bridge the gap between complex machine learning models and biological insights.

Main Methods:

• Introduction of Positional Oligomer Importance Matrices (POIMs).
• Development of an efficient algorithm for POIM computation.
• Utilizing k-mer feature correlations, including overlapping k-mers.

Main Results:

• POIMs provide a method to interpret SVM-based sequence classification.
• POIMs account for the correlation structure of k-mer features.
• POIMs generalize and visualize sequence patterns relevant to biological phenomena, akin to sequence logos.

Conclusions:

• POIMs significantly improve the interpretability of SVM sequence classifiers.
• This method facilitates the connection of machine learning findings to biological knowledge.
• POIMs offer a powerful tool for analyzing and understanding biological sequence data.