Memo: Integration methods • SeuratIntegrate

SeuratIntegrate provides access to R and Python methods to correct batch effect:

Comprehensive overview of the integration methods provided by SeuratIntegrate
Method	Implementation	Underlying algorithm	Reference
ComBat	sva (Bioconductor)	Empirical Bayes adjustment	Johnson et al., 2007
Harmony	harmony (CRAN)	Iterative embedding correction	Korsunsky et al., 2019
MNN	batchelor (Bioconductor)	Mutual Nearest Neighbors	Haghverdi et al., 2018
BBKNN	bbknn (GitHub)	Batch-balanced nearest neighbors	Polański et al., 2020
scVI	scvi-tools (GitHub)	Variational autoencoder	Lopez et al., 2018
scANVI	scvi-tools (GitHub)	Semi-supervised variational autoencoder	Xu et al., 2021
Scanorama	scanorama (GitHub)	Manifold alignment	Hie et al., 2019
trVAE	scArches (GitHub)	Conditional variational autoencoder	Lotfollahi et al., 2020

SeuratIntegrate provides a new interface to integrate the layers of an object: DoIntegrate(). Moreover, SeuratIntegrate is compatible with CCA and RPCA (included in Seurat) and FastMNN (from SeuratWrappers)

The table below summarizes the integrations methods you can run with DoIntegrate() along with the expected inputs and produced outputs of each method.

Comprehensive overview of inputs and outputs of the integration methods
Method	Package	Function	Input	Output(s)
ComBat	SeuratIntegrate	`CombatIntegration()`	Layer	Layer with corrected counts
Harmony	SeuratIntegrate	`HarmonyIntegration()`	PCA	DimReduc (corrected PCA)
MNN	SeuratIntegrate	`MNNIntegration()`	Layer	Layer with corrected counts
BBKNN	SeuratIntegrate	`bbknnIntegration()`	PCA	Graphs (edge weights can be distances or connectivities)
scVI	SeuratIntegrate	`scVIIntegration()`	Layer	DimReduc (latent space)
scANVI	SeuratIntegrate	`scANVIIntegration()`	Layer	DimReduc (latent space)
Scanorama	SeuratIntegrate	`ScanoramaIntegration()`	Layer	Layer with corrected counts DimReduc (corrected embedding)
trVAE	SeuratIntegrate	`trVAEIntegration()`	Data layer (`recon.loss = "mse"`) Counts layer (otherwise)	DimReduc (latent space)
CCA	Seurat	`CCAIntegration()`	PCA	DimReduc (corrected PCA)
RPCA	Seurat	`RPCAIntegration()`	PCA	DimReduc (corrected PCA)
FastMNN	SeuratWrappers	`FastMNNIntegration()`	Layer	Layer with corrected counts DimReduc (corrected PCA)