Comparing two K-category assignments by a K-category correlation coefficient.

Area of Science:

• Bioinformatics
• Computational Biology
• Sequence Analysis

Background:

• Current biological sequence prediction evaluations often use the Matthews correlation coefficient, limited to binary classification.
• This binary approach leads to information loss when dealing with multi-category predictions, such as RNA secondary structure.

Purpose of the Study:

• To propose an extended correlation coefficient applicable to K-category classification problems.
• To demonstrate its utility in evaluating RNA secondary structure predictions, which can involve three categories (paired, unpaired, unknown).

Main Methods:

• Development of a generalized correlation coefficient for K-category assignments.
• Application and validation of the K-category coefficient for RNA secondary structure prediction evaluation.
• Comparison with existing performance measures for protein secondary structure prediction.

Main Results:

• The proposed K-category correlation coefficient effectively evaluates predictions with more than two categories.
• It shows high applicability for RNA secondary structure prediction, accommodating 'unknown' or unreliable predicted pairs.
• The measure demonstrates strong agreement with established metrics for protein secondary structure prediction ranking.

Conclusions:

• The K-category correlation coefficient offers a more informative and accurate method for evaluating multi-category biological sequence predictions.
• This new metric enhances the assessment of RNA secondary structure prediction accuracy.
• Software and server are available for broader application and research.