Forest classification trees and forest support vector machines algorithms: Demonstration using microarray data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-dimensional data in -omics studies presents classification challenges.
• Dimensionality reduction is crucial for accurate multi-class classification.

Purpose of the Study:

• To introduce two novel algorithms, forest classification tree (FCT) and forest support vector machines (FSVM), for multi-class classification in high-dimensional -omics data.
• To demonstrate the efficacy of FCT and FSVM in identifying important variables and improving classification accuracy.

Main Methods:

• Forest Classification Tree (FCT): Randomly selects variables, grows classification trees (CTs), and constructs a final tree using frequent variables from low misclassification trees.
• Forest Support Vector Machines (FSVM): Replaces CTs with Support Vector Machines (SVMs) within the forest framework.
• Application: Utilized prostate gene expression data for classifying four tumor types.

Main Results:

• FCT achieved perfect classification (AMR=0) with 100 markers at a threshold of 0.001.
• At a threshold of 0.01, FCT used 15 markers for classification with 7% AMR.
• FSVM demonstrated reduced misclassification rates by adjusting the fraction of the forest used for the classifier.

Conclusions:

• FCT and FSVM are effective methodologies for variable identification in high-dimensional datasets.
• FCT offers insights into data structure and provides interpretable decision rules.
• These algorithms offer robust solutions for complex classification tasks in biological research.