Batch effect correction methods (Seurat v3, Harmony, fastMNN, Liger)

Question

[re_raz]

I applied different batch effect correction methods including Seurat v3 integration, Harmony, fastMNN, and Liger on 52 single-cell RNA PBMC samples from different 4 public datasets. The results from the methods presented in the figure (samples grouped by sample_ID, datasets, SingleR annotation)

I read the paper A benchmark of batch-effect correction methods for single-cell RNA sequencing data compared the batch effect methods, however, I am still confused about what method I use.

Are there metrics to compare results?

Any advice on which method is the best in my case?

thanks in advance

[antonioggsousa]

Just a comment rather than an answer.

The Theis lab published a more recent benchmarking (please see the paper) where they provided the scBI python package with several metrics, some known and others new, to assess the batch removal and retention of biological signal.

You can find the scBI python package in their github: https://github.com/theislab/scib

The application of some of these metrics may help you to justify/support your decision.

I hope this helps,

António