HarmonizR: blocking and singular feature data adjustment improve runtime efficiency and data preservation.

Area of Science:

• Multi-omics data analysis
• Bioinformatics
• Statistical genetics

Background:

• Data adjustment is crucial for omics data analysis (e.g., single-cell RNA, proteomics).
• Data integration introduces batch effects and missing values, hindering standard adjustment algorithms like ComBat and limma.
• The HarmonizR framework offers missing value-tolerant batch effect adjustment for omics data.

Purpose of the Study:

• To significantly improve the HarmonizR framework for batch effect adjustment in omics data.
• To address limitations in runtime and feature retention of the original HarmonizR.
• To enhance the robustness and efficiency of batch effect reduction, particularly for rare tumor entities.

Main Methods:

• Introduction of a novel blocking strategy to reduce runtime and support parallel processing.
• Integration of a "unique removal" strategy to preserve more features during adjustment.
• Testing and validation on small and large real-world omics datasets.

Main Results:

• Severely improved runtime for HarmonizR on both small and large datasets.
• Demonstrated enhanced feature retention, with up to 103.9% feature rescue in tested datasets.
• Validation of improved performance for batch effect reduction in integrated omics data.

Conclusions:

• The updated HarmonizR addresses previous shortcomings, offering substantial runtime improvements beyond basic parallelization.
• The enhanced feature rescue capability improves the algorithm's ability to retain valuable features during batch effect adjustment.
• The improved HarmonizR provides a quicker and more robust solution for batch effect reduction in complex, integrated omics datasets.