Hello, I am running single cell seq analysis and am being held at the IntegrateData step: seurat_integrated <- IntegrateData(anchorset = immune_anchors, normalization.method = "SCT")
it's a problem of memory . I tried to reduce memory by applyin reciprocal PCA when I run FindIntegrationAnchors but it didn't solve the problem. Is there other hints ?
otherwise can I still apply the old protocol NormalizeData, FindVariableFeatures, and ScaleData instead of SCTransform and still have the same output ( I already applied the old protocol and i didn't face memory problem) that why I'm wondering if it's really worth it to apply SCTransform and if the outputs differs between the 2 approaches ?
finally what's the advantage of SCTransform compared to the old protocol ?
How much memory do you have available? Find find Harmony to give reasonable results in my hand with very limited memory footprint.
60GB otherwise if it doesn't work .can I still apply the old protocol NormalizeData, FindVariableFeatures, and ScaleData instead of SCTransform and still have the same output ( I already applied the old protocol and i didn't face memory problem) that why I'm wondering if it's really worth it to apply SCTransform and if the outputs differs between the 2 approaches ?
finally what's the advantage of SCTransform compared to the old protocol ?
sctransforms claims to better regularize and variance-stabilize the data but as a non-Seurat user I have integrated heterogeneous samples and those with strong confounders (e.g. homeostasis vs acute inflammation) with harmony before and this works very decently. You don't need (in terms of being dependent on it) sctransform, it's one of many options.