HighDimMixedModels.jl: Robust high-dimensional mixed-effects models across omics data.

Area of Science:

• Statistical modeling
• Bioinformatics
• Computational biology

Background:

• High-dimensional mixed-effects models are crucial for analyzing clustered data where covariates exceed samples.
• Penalized likelihood methods with coordinate descent are common but may not guarantee global optima.
• Omics data (transcriptome, GWAS, microbiome) present unique challenges for these models.

Purpose of the Study:

• To empirically investigate the coordinate descent algorithm's behavior in high-dimensional mixed-effects models for omics data.
• To compare the performance of SCAD and LASSO penalties in variable selection and estimation accuracy.
• To provide a practical tool for researchers to implement SCAD penalty fitting.

Main Methods:

• Empirical study using simulated and real transcriptome, genome-wide association, and microbiome data.
• Comparison of smoothly clipped absolute deviation (SCAD) and least absolute shrinkage and selection operator (LASSO) penalties.
• Implementation of a Julia package (HighDimMixedModels.jl) for fitting models with the SCAD penalty.

Main Results:

• Simulations offered new insights into the coordinate descent algorithm's performance in omics data settings.
• SCAD penalty demonstrated superior performance over LASSO in both variable selection and estimation accuracy.
• The HighDimMixedModels.jl package facilitates the application of SCAD penalty models.

Conclusions:

• The SCAD penalty is a more effective choice than LASSO for high-dimensional mixed-effects models in omics data analysis.
• The developed Julia package empowers researchers to utilize advanced statistical methods for biological data.
• This work contributes to improved analytical techniques for complex biological datasets.