PLSDAbatch

PLSDAbatch corrects for batch effects prior to downstream analysis in microbiome data by estimating treatment and batch variation with latent components and subtracting batch-associated components while preserving biological variation of interest.

Key Features:

• Non-parametric multivariate correction: PLSDA-batch is non-parametric, multivariate, and allows for ordination and data visualisation.
• Centered log-ratio transformation integration: Combined with centered log-ratio transformation to address uneven library sizes and compositional structure.
• Unbalanced design variant: A variant for unbalanced batch x treatment designs commonly encountered in studies with small sample sizes.
• Discriminative variable selection variant: A variant for selection of discriminative variables amongst treatment groups to avoid overfitting in classification problems.

Scientific Applications:

• Microbiome data batch effect correction: Corrects for batch effects prior to any downstream analysis in microbiome data.
• Compositional data analysis: Addresses characteristics of microbiome data including uneven library sizes and compositional structure that existing correction methods have ignored.
• Small sample size studies: Applicable to studies with small sample sizes and unbalanced batch x treatment designs.
• Classification problems: Applicable to classification problems requiring selection of discriminative variables amongst treatment groups.

Methodology:

PLSDA-batch first estimates treatment and batch variation with latent components, then subtracts batch-associated components from the data whilst preserving biological variation of interest. The method is based on Projection to Latent Structures Discriminant Analysis and is combined with centered log-ratio transformation.