ULV: A robust statistical method for clustered data, with applications to multi-subject, single-cell omics data

Area of Science:

• Genomics and Proteomics
• Computational Biology
• Single-cell Analysis

Background:

• Technological advancements enable high-resolution biological measurements, including single-cell profiling.
• Analyzing single-cell data presents challenges like small sample sizes, non-normality, dropouts, and outliers.
• Existing methods may struggle with the complexities of multi-omic single-cell datasets.

Purpose of the Study:

• To introduce a novel computational method, U-statistic based latent variable (ULV), for analyzing complex single-cell data.
• To address limitations of current methods in handling small sample sizes, non-normality, and other data challenges.
• To provide a flexible and computationally feasible framework for both single-cell RNA and protein abundance data.

Main Methods:

• Developed a U-statistic based latent variable (ULV) framework.
• Leveraged the robustness of rank-based statistics and efficiency of parametric methods for small sample sizes.
• Designed a computationally feasible method to simultaneously address data limitations like dropouts and outliers.

Main Results:

• ULV controls false positives at desired significance levels.
• Demonstrated effectiveness in single-cell proteomics (AML) and single-cell RNA (COVID-19) studies.
• ULV identified differentially expressed proteins and genes missed by traditional methods, including those affected by covariates and less biased by high expression levels.

Conclusions:

• ULV offers a robust and flexible approach for single-cell data analysis, outperforming existing methods.
• The method successfully identified novel biological insights in both cancer and infectious disease studies.
• ULV provides a valuable tool for advancing our understanding of biological mechanisms at the single-cell level.