Sparse Semiparametric Discriminant Analysis for High-dimensional Zero-inflated Data

Area of Science:

• Bioinformatics
• Statistical Genetics
• Computational Biology

Background:

• Sequencing technologies generate high-dimensional biological data with inherent skewness and zero-inflation.
• Linear classification methods face challenges due to violated distribution assumptions in such data.
• Existing data transformation methods introduce ambiguity and affect model performance.

Purpose of the Study:

• To propose a novel semiparametric framework for discriminant analysis robust to data characteristics.
• To address skewness and zero inflation in high-dimensional biological data.
• To improve classification accuracy and interpretability for sequencing-based data.

Main Methods:

• Developed a semiparametric framework using a truncated latent Gaussian copula model.
• Incorporated L1 sparsity regularization for enhanced model interpretability.
• Established theoretical consistency of classification directions in high-dimensional settings.

Main Results:

• The proposed model effectively handles skewed and zero-inflated data.
• Demonstrated robustness against various data transformation methods.
• Achieved superior classification accuracy compared to existing approaches.

Conclusions:

• The novel framework offers a robust and interpretable solution for discriminant analysis of high-dimensional biological data.
• The method shows promise for applications in microbiome, cancer genomics, and single-cell RNA sequencing.
• This approach overcomes limitations of traditional methods when dealing with complex biological datasets.