Score matching for differential abundance testing of compositional high-throughput sequencing data.

Area of Science:

• Statistical modeling
• Bioinformatics
• Genomics

Background:

• Sparse compositional count data, common in high-throughput sequencing, often exhibit excess zeros.
• Existing power interaction models handle zeros but lack covariate integration for heterogeneous populations.
• Differential abundance (DA) testing is crucial for identifying biological differences but can be confounded by correlated features.

Purpose of the Study:

• To extend power interaction models for covariate inclusion in heterogeneous populations.
• To develop a novel differential abundance testing scheme, cosmoDA, robust to correlated features.
• To provide a framework linking transformations to DA results and assessing their impact.

Main Methods:

• Extension of a-b power interaction models to incorporate covariate information.
• Development of cosmoDA, a DA testing scheme using generalized score matching estimation.
• Benchmarking on simulated and real high-throughput sequencing data.

Main Results:

• cosmoDA accurately estimates feature interactions in heterogeneous populations.
• The method significantly reduces the false discovery rate in DA testing of correlated features.
• cosmoDA explicitly links transformations to DA results, enabling impact assessment.

Conclusions:

• cosmoDA offers a robust approach for differential abundance analysis of sparse count data with covariates.
• The method improves accuracy and reduces false positives, particularly in complex biological datasets.
• cosmoDA provides valuable insights into the influence of data transformations on downstream analyses.