Integration methods implemented in panpipes

The panpipes integration workflow implements a variety of tools to batch correct individual modalities and/or integrate across modalities to produce a reduced dimension representation of the dataset.
There are different tools available for each modality: RNA (also referred to as GEX), PROT (can be referred to as ADT) and ATAC which can be integrated into any preferred combination customising the integration workflow configuration file and running panpipes integration make full. After the results of the integration are inspected, the final object is created with panpipes integration make merge_integration.

The ideal way to run panpipes integration is to use the output MuDatafile from panpipes preprocess since it will already be in the required format. However, if using independent MuData the object should contain normalised data in the X slot of each modality, a ‘raw_counts’ layer in each modality, and a sample_id column in each slot of the obs and the outer obs.

Users can choose which integration method they want to apply based on their experiment, their experience with the tools or available benchmarks: we link all the relevant resources below.

We don’t believe in “one method fits all”, we instead offer panpipes as a framework to run multiple tools efficiently, ensuring reproducibility of results. We believe this will empower users to choose the method that best fits their biological question, keeping a record of the hyperparameters in the configuration files, so you can safely re-run your analysis and share it with collaborators. We will continue to update the integration methods offered in panpipes and we invite you to contribute yours! The following table describes the different methods currently supported and their specificities:

Method	type of integration	modalities	code	references	benchmarks paper
harmony	unimodal (batch correction)	rna, atac, prot	harmony	Korsunsky  et al 2019	Luecken et al 2022
BBKNN	unimodal (batch correction)	rna, atac, prot	BBKNN	Polański et al 2020	Luecken et al 2022
Scanorama	unimodal (batch correction)	rna	Scanorama	Hie, Bryson, and  Berger 2019	Luecken et al 2022
scVI	unimodal (batch correction)	rna	scVI	Gayoso et al 2022	Luecken et al 2022
MultiVI	multimodal	atac, rna	MultiVI	Ashuach et al 2023	Lee, Kaestner, and Li 2023
totalVI	multimodal	prot, rna	totalVI	Gayoso  et al. 2021	Makrodimitris et al 2024
MOFA	multimodal	rna, atac, prot	MOFA	Argelaguet et al 2020	Lee, Kaestner, and Li 2023
WNN	multimodal	rna, atac, prot	WNN	Hao et al 2021	Lee, Kaestner, and Li 2023