Ordinal response prediction using bootstrap aggregation, with application to a high-throughput methylation data set.

Area of Science:

• Genomics
• Bioinformatics
• Translational Research

Background:

• High-throughput genomic experiments generate high-dimensional data.
• Translational research often involves predicting inherently ordinal outcomes.
• Existing nominal classification methods may lose information when applied to ordinal data.

Purpose of the Study:

• To examine the effectiveness of alternative ordinal splitting functions combined with bootstrap aggregation for classifying ordinal response data.
• To identify methods that preserve information in ordinal outcomes for improved classifier performance.

Main Methods:

• Investigated alternative ordinal splitting functions (ordinal impurity, ordered twoing) with bootstrap aggregation.
• Applied methods to high-dimensional genomic and methylation data sets.
• Compared performance against previously described methods for ordinal response classification.

Main Results:

• Ordinal impurity and ordered twoing methods demonstrated desirable properties for ordinal response classification.
• These ordinal methods performed well in comparison to other approaches.
• The effectiveness of ordinal ensemble methods was demonstrated on a high-throughput methylation data set.

Conclusions:

• Ordinal splitting functions, specifically ordinal impurity and ordered twoing, are effective for classifying ordinal response data.
• These methods offer improved predictive performance by leveraging the inherent order of outcomes.
• Ordinal ensemble methods are valuable for developing multigenic classifiers in high-throughput genomic studies.