Generating correlated data for omics simulation.

Area of Science:

• Computational Biology
• Bioinformatics
• Statistical Modeling

Background:

• Realistic omics data simulation is vital for benchmarking computational pipelines.
• Omics data often exhibits correlations between measured features, which are frequently ignored in simulations due to computational challenges.
• Existing simulation methods struggle to incorporate feature dependencies efficiently.

Purpose of the Study:

• To introduce efficient methods for generating omics-scale data with correlated measures.
• To demonstrate the impact of including correlations in benchmarking studies.
• To provide a flexible R package for simulating dependent omics data.

Main Methods:

• Developed three simulation approaches based on a Gaussian copula model.
• Utilized a covariance matrix decomposition (diagonal and low-rank) for computational efficiency.
• Implemented methods in the R package `dependentsimr` for various marginal distributions.

Main Results:

• Showed increased variance in DESeq2 results when feature dependence was included.
• Demonstrated performance improvements for CYCLOPS in inferring circadian time with gene-gene dependencies.
• Validated the importance of correlation in omics data simulation for accurate benchmarking.

Conclusions:

• Efficient simulation of correlated omics data is achievable and essential for robust benchmarking.
• Incorporating feature dependencies can significantly impact the performance of omics analysis tools.
• The `dependentsimr` package offers a practical solution for generating realistic, dependent omics data.