Table 3:

Intra-modality data harmonization approaches

Method name	Strategy	Main advantages	Main limitations	Citation
ComBat	Bayesian empirical	Removes batch effect in most cases	Removes biological signal in most cases	[4]
RUV	Linear model	Effective with spike-in controls	Individual variants make specific assumptions about the data	[79]
removeBatchEffect	Linear model	Generalizable to most transcriptomic data types	May be less effective in complex experimental designs	[80]
SVN	Linear model	Generalizable to many cases	Assumes that feature similarities between datasets are due to biology	[81]
mnnCorrect	Mutual nearest neighbours	Accounts for heterogeneity within sample groups	Restricted to single-cell data	[78]
MINT	Multivariate model	Robust to overfitting and strong multidimensional technical variation	Minimum sample count requirement	[82]
Scanorama	Mutual nearest neighbours	Scales to very large sample sizes. Robust to overcorrection	Restricted to single-cell data	[83]
MultiCluster	Tensor decomposition	Accounts for multiple batch variables simultaneously	Restricted to 3-way variable comparisons	[84]
zeroSum	Zero sum regression	Generalizable across different technologies and platforms	Weak or non-linear features may be masked by strong features	[85]

Batch is a special case of intra-modality harmonization and is included for completeness because many underlying strategies used are applicable to broader data integration. All methods are restricted to a single data modality of transcriptomics. Names, strategies, advantages, and limitations of each method are provided. Regarding advantages and limitations, a few major points are highlighted. A citation for reference to the original publication of each method is provided where full details can be obtained.