MetaDICT

MetaDICT integrates microbiome data by removing batch effects and preserving biological variation across heterogeneous datasets while avoiding overcorrection in the presence of unobserved confounding variables.

Key Features:

• Batch effect estimation via causal inference weighting: MetaDICT initially estimates batch effects using weighting methods from causal inference literature.
• Shared dictionary learning refinement: MetaDICT refines batch effect estimation through novel shared dictionary learning.
• Overcorrection avoidance: MetaDICT better avoids overcorrection of batch effects when unobserved confounding variables exist, datasets are highly heterogeneous across studies, or batch is completely confounded with some covariates.
• Multi-level comparable embedding generation: MetaDICT generates comparable embeddings at both taxa and sample levels to reveal hidden structure of integrated data.

Scientific Applications:

• Microbial interaction characterization: MetaDICT enables characterization of microbial interactions through integrative analysis.
• Generalizable microbial signature identification: MetaDICT identifies microbial signatures that generalize across studies.
• Outcome prediction enhancement: MetaDICT enhances accuracy of outcome prediction in integrative studies.
• Colorectal cancer metagenomics integration: MetaDICT supports integrative analysis of colorectal cancer metagenomics studies.
• Immunotherapy microbiome meta-analysis: MetaDICT enables meta-analysis of immunotherapy microbiome studies.

Methodology:

MetaDICT initially estimates batch effects by weighting methods in causal inference literature, then refines the estimation via novel shared dictionary learning.