Debiased high-dimensional regression calibration for errors-in-variables log-contrast models.

Area of Science:

• Statistics
• Bioinformatics
• Computational Biology

Background:

• Analyzing high-dimensional microbiome and metagenomic data presents challenges due to measurement errors in compositional covariates.
• Existing statistical models often struggle with the complexities of mismeasured or contaminated compositional data.

Purpose of the Study:

• To develop a statistical inference method for high-dimensional compositional data affected by measurement errors.
• To introduce a novel calibration approach for linear log-contrast models in the context of microbiome and metagenomic data analysis.

Main Methods:

• Developed a calibration approach specifically for the linear log-contrast model.
• Established asymptotic normality of the estimator under sparse parameter conditions.
• Utilized numerical experiments and a real-world microbiome study for validation.

Main Results:

• The proposed high-dimensional calibration strategy effectively minimizes bias in compositional data analysis.
• Achieved expected coverage rates for confidence intervals, enhancing statistical inference reliability.
• Demonstrated the method's efficacy in a microbiome study context.

Conclusions:

• The novel calibration method provides a robust solution for statistical inference on high-dimensional compositional data with measurement errors.
• The methodology shows broad applicability beyond microbiome studies, adaptable to various research fields.
• This work pioneers statistical inference techniques for contaminated high-dimensional compositional datasets.