Pathway Lasso: Pathway Estimation and Selection with High-Dimensional Mediators.

Area of Science:

• Statistics
• Computational Biology
• Neuroscience

Background:

• Delineating mediation pathways with numerous genetic and brain mediators is crucial but challenging.
• Traditional structural equation modeling (SEM) struggles with high-dimensional data, leading to instability and computational issues.

Purpose of the Study:

• To propose a sparse mediation model using regularized SEM for high-dimensional mediators.
• To introduce the novel 'Pathway Lasso' penalty for simultaneous estimation and selection of mediation effects.

Main Methods:

• Developed a regularized SEM approach with the Pathway Lasso penalty.
• Implemented a fast ADMM-type algorithm for parameter computation with closed-form updates.
• Proved the asymptotic consistency of the Pathway Lasso estimator.

Main Results:

• The proposed sparse mediation model effectively handles high-dimensional mediators.
• Pathway Lasso enables computationally tractable optimization for estimating and selecting pathway effects.
• Demonstrated superior pathway selection accuracy and reduced estimation bias compared to existing methods on simulated and fMRI data.

Conclusions:

• The Pathway Lasso approach offers a robust and efficient solution for sparse mediation analysis in high-dimensional settings.
• This method enhances the ability to identify significant mediation pathways in complex biological and genetic studies.
• The developed algorithm provides a computationally efficient tool for researchers in statistics and neuroscience.