SPReM: Sparse Projection Regression Model For High-dimensional Linear Regression.

Area of Science:

• Statistics
• Genetics
• Bioinformatics

Background:

• High-dimensional data presents challenges for standard statistical methods.
• Existing approaches like Hotelling's T-squared test lack statistical power.
• Multivariate regression with numerous responses requires advanced modeling.

Purpose of the Study:

• To develop a sparse projection regression modeling (SPReM) framework.
• To address low statistical power in high-dimensional multivariate regression.
• To enable simultaneous dimension reduction, response selection, estimation, and testing.

Main Methods:

• Proposed two novel heritability ratios for multivariate response analysis.
• Formulated estimation as a sparse unit rank projection (SURP) problem.
• Developed a fast optimization algorithm for SURP and extended it to sparse multi-rank projection (SMURP).

Main Results:

• SPReM effectively performs dimension reduction and response selection.
• The framework accounts for correlations among multivariate responses.
• SPReM demonstrated superior performance compared to state-of-the-art methods in simulations and real data analysis.

Conclusions:

• SPReM offers a powerful and effective approach for high-dimensional multivariate regression.
• The proposed methods overcome limitations of traditional statistical techniques.
• SPReM provides a robust framework for genetic and bioinformatics applications.