Synthetic Data Generation and Nonparametric Techniques for Assessing Multivariate Similarity to Address Small-Sample...

Area of Science:

• Biomedical data science
• Statistical modeling
• Computational biology

Background:

• Biomedical research frequently faces small-sample data limitations, impacting study outcomes.
• Synthetic data generation offers a potential solution, but its fidelity requires rigorous evaluation.
• Existing methods struggle to generate adequate synthetic data for high-dimensional, low-sample scenarios.

Purpose of the Study:

• To evaluate a previously proposed synthetic data generator for biomedical applications.
• To develop and apply a nonparametric method for assessing multivariate similarity using the Cramér-Wold theorem and random projection testing.
• To investigate the conditions under which bivariate correlation absence approximates independence in non-normal distributions and evaluate data compression artifacts.

Main Methods:

• Developed a nonparametric multivariate similarity assessment based on Cramér-Wold theorem and random projection testing.
• Investigated the approximation of independence by bivariate correlation absence in non-normal settings.
• Evaluated artifacts introduced by data compression during synthetic data generation.
• Established a formal testing framework using Bernoulli trials, aggregated outcomes, and a standardized normal test-statistic.

Main Results:

• The synthetic generator produced high-fidelity multivariate synthetic data when bivariate correlation approximated independence in non-normal settings.
• In highly compressed data, residual modes were best modeled as normally distributed, irrespective of their intrinsic form.
• The developed projection framework effectively evaluated the full multivariate covariance structure.

Conclusions:

• The evaluated synthetic data generator shows promise for generating high-fidelity data in specific non-normal, low-sample regimes.
• The nonparametric method for assessing multivariate similarity is scalable and adaptable for evaluating synthetic data quality.
• Further research is needed to apply these methods to higher-dimensional and diverse biomedical datasets.