Batch correction preserving biological difference

Question

[foejvs546]

Hi everyone,

I have four samples of scRNA-seq data. Two come from tumour tissue, two from healthy tissue. I want to identify DEGs between cancer and healthy cells.

Can I batch correct for sample (i.e., sample_1_cancer, sample_2_cancer, sample_3_healthy, sample_4_healthy), without affecting any biological difference due to sample type (healthy or cancer)?

Thank you!

Answer 1

[OmnibusX]

While analyzing single-cell datasets, in my lab, we use batch-corrected results for visualization purposes (t-SNE/UMAP) or labeling cell types based on clusters. In DEG analysis, the fold-change between groups requires raw counts for precise calculation. To avoid batch effects, we usually choose to analyze each cell type at a time, and each group for comparison is chosen from multiple batches. For example, B cells from sample_1_cancer and sample_2_cancer versus B cells from sample_3_healthy and sample_4_healthy. This approach reduces the significance of a single batch effect within the comparison group.

I hope this can be helpful for you.

[foejvs546]

Thank you for your help OmnibusX!

So if I understand, the batch correction is for visualisation, then differential analysis is performed on non-batch corrected counts? If you have time to reply, I'd love to know what method you use for differential analysis. Thanks.

[OmnibusX]

Yes, your understanding is correct.

For your question, typically, we use a t-test for differential analysis. However, when dealing with data combined from very different batches (such as different technologies or labs), we prefer to use the Wilcoxon rank sum test. This method uses ranks and is particularly useful for minimizing effects arising from variations in data scales between batches.

If you want to quickly try those methods on your dataset, you can use our app here: https://omnibusx.com/apps.

[foejvs546]

Thank you OmnibusX, I appreciate it