qtQDA: quantile transformed quadratic discriminant analysis for high-dimensional RNA-seq data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Gene expression data from RNA-sequencing (RNA-seq) is crucial for modern medicine.
• Existing classification methods may not fully account for the inherent dependencies between gene measurements.

Purpose of the Study:

• To introduce a novel classification method for RNA-seq data.
• To address the challenge of dependent measurements in gene expression data.

Main Methods:

• Proposed a new statistical model for gene expression data.
• Data is marginally negative binomial but dependent.
• Employed a quantile transformation (qt) followed by Gaussian quadratic discriminant analysis (QDA) with regularized covariance matrices, termed qtQDA.

Main Results:

• qtQDA demonstrated excellent performance on real-world datasets.
• The method showed advantages over several existing classification approaches.
• An R package for qtQDA is publicly available.

Conclusions:

• qtQDA is a powerful new tool for gene expression classification.
• The method effectively incorporates gene dependencies, improving classification accuracy.
• This approach has significant potential for applications in precision medicine.