MoDaH achieves rate optimal batch correction.

Area of Science:

• Computational biology
• Statistical genomics
• Bioinformatics

Background:

• Batch effects are a major challenge in single-cell omics data analysis.
• Existing computational methods for batch correction lack theoretical guarantees for reliability and generalization.

Purpose of the Study:

• Introduce Mixture-Model-based Data Harmonization (MoDaH), a statistically principled batch correction algorithm.
• Provide the first theoretical guarantees for batch correction methods.

Main Methods:

• Developed a Gaussian-mixture-model with explicit batch effect parametrization.
• Established minimax optimal error rates for batch correction.
• Proved MoDaH achieves optimal rates using advances in clustering anisotropic Gaussian mixtures.

Main Results:

• MoDaH achieves theoretical optimality in batch correction.
• Demonstrated comparable or superior empirical performance to state-of-the-art methods (Harmony, Seurat-V5, LIGER).
• Successfully balanced technical noise removal with biological signal preservation across diverse datasets.

Conclusions:

• MoDaH is the first batch correction algorithm with theoretical guarantees.
• Offers a reliable and effective approach for single-cell omics data harmonization.
• Advances the field of computational biology by providing a principled method for batch effect correction.