Processing-bias correction with DEBIAS-M improves cross-study generalization of microbiome-based prediction models.

Area of Science:

• Microbiome research
• Bioinformatics
• Computational biology

Background:

• Microbiome profiling protocols introduce processing biases affecting microbial detection.
• These biases hinder the identification of biologically interpretable and generalizable signals.
• Existing computational batch-correction methods are often non-interpretable and prone to overfitting.

Purpose of the Study:

• To present DEBIAS-M, an interpretable framework for microbiome data processing bias inference and correction.
• To facilitate domain adaptation across diverse microbiome studies.
• To improve the accuracy and generalizability of microbiome data analysis.

Main Methods:

• DEBIAS-M employs domain adaptation with phenotype estimation and batch integration.
• It learns microbe-specific bias-correction factors for each batch.
• These factors minimize batch effects while maximizing cross-study associations with phenotypes.

Main Results:

• DEBIAS-M demonstrated improved cross-study prediction accuracy compared to existing methods across various benchmarks (16S rRNA, metagenomics, classification, regression).
• Inferred bias-correction factors were stable, interpretable, and linked to experimental protocols.
• The framework enhances the modeling of microbiome data and identifies generalizable signals.

Conclusions:

• DEBIAS-M offers an interpretable solution for correcting processing biases in microbiome studies.
• It enables more reliable identification of biological signals that generalize across different studies and experimental conditions.
• The framework advances microbiome data analysis by improving model performance and interpretability.