qtQDA

qtQDA applies quantile-transformed quadratic discriminant analysis to classify high-dimensional RNA-seq gene expression data by modeling marginally negative binomial counts with gene dependencies to improve covariance estimation and classification accuracy.

Key Features:

• Dependence Modeling: Models gene expression as marginally negative binomial with dependencies to represent inter-gene correlation.
• Quantile Transformation (qt): Applies a quantile transformation to RNA-seq counts to stabilize variance and produce a Gaussian-like distribution.
• Gaussian Quadratic Discriminant Analysis (QDA): Performs Gaussian QDA using regularized covariance matrix estimates to classify samples based on mean and covariance differences.
• Covariance Matrix Estimation: Estimates covariance matrices using a local dependence function, providing more accurate estimates than Maximum Likelihood Estimators (MLE) as reported.
• Performance on RNA-seq: Demonstrates improved classification performance and reduced error rates on real RNA-seq datasets, notably in cancer prediction tasks.

Scientific Applications:

• Oncology / Cancer prediction: Classification of cancer types and prediction in oncology using RNA-seq gene expression profiles.
• Biomarker discovery and personalized medicine: Support for biomarker discovery and personalized medicine through improved classification of RNA-seq expression patterns.

Methodology:

The method models RNA-seq counts as marginally negative binomial with dependencies, applies a quantile transformation to obtain a Gaussian-like distribution, estimates regularized covariance matrices via a local dependence function, and performs Gaussian quadratic discriminant analysis for classification.