Any way to use pre-computed cell embeddings as input for dyn.tl.moments?

[altairwei]

In scvelo, I can use the use_rep argument of scvelo.pp.moments to compute first-/second-order moments based on the given cell embeddings. Many batch effect correcting tools are operated on dimensionality reduction space, such as harmony or fastMNN. It's important for multi-sample scRNA-Seq data to carry out downstream analysis based on batch-corrected cell embeddings.

However, preprocessing steps in dynamo are wrapped within one function recipe_monocle, there is no way to break it down and plug my batch-corrected cell embeddings in it. Is there any workaround fot the problem?

[Xiaojieqiu]

hey @altairwei , you can usedyn.tl.moments(adata, X_data=adata.obsm['X_harmony'])or others other embedding in .obsm to calculate first/second and mix-order moments with dynamo.

Dynamo's preprocessing module is actually very flexible and we have integrated Pearson residual and harmony in our model. Unfortunately, I am busy with job searches these days, but you can check this notebook to get it started: https://github.com/aristoteleo/dynamo-tutorials/blob/master/tutorial_preprocess.ipynb

We will have a tutorial on preprocessing in our readthedocs page sometime next month.

[altairwei]

Hi @Xiaojieqiu, tutorial_preprocess.ipynb is great, and it made me realize that dynamo's preprocessing is indeed customizable. I am interested in dynamo's negbin mdoe for conventional scRNA-seq, but I just want to normalize spliced&unspliced data then calculate moments based on harmony cell embeddings (the customized preprocessor part is not yet written in the tutorial). However, the source codes of dyn.tl.dynamics are complex, and I can't find a way to separate the logic of moments calculation from it. That makes it seem unlikely to calculate moments based on given cell embeddings. Could you give me more hints?

[Xiaojieqiu]

please see my above code on running dyn.tl.moments. Once moments calculated before dynamics, it won't be recomputed

[altairwei]

Gosh, I totally didn't realize that! Thank you so much!