The code to preprocess data

[SiaGuo]

Hi,

The STAMP is quiet an efficient algorithm to integrate multi samples. But I wonder how to preprocess data. For example, I have four samples generated by 10X Visium, and how can I preprocess the data to the format as the input for STAMP. I would like to know how this part of the code is implemented. Thank you for your help!

Sincerely,
Sia G

[Chengwei94]

@SiaGuo,

You can just use the usual scanpy preprocess. However, the input into the algorithm is counts data, so rmb to save the counts data in the layer = "counts". Then for multiple samples, you can set the categorical_covariate_keys= [your_batch]. We will update our docs soon to make it more clear.

[SiaGuo]

Got it! Thanks.

[katimbach]

Hi! I have a follow-up question regarding this. If I have a similar situation, with multiple samples that I would like to combine, but also want to account for the spatial neighbors, as in the mouse brain example (https://jinmiaochenlab.github.io/scTM/notebooks/stamp/example2/), is this possible?

I'm a bit confused, as it seems in the mouse brain tutorial the data is explicitly provided as a covariate, whereas in other tutorials, such as that with lung cancer, the neighbor graph is created but not provided as a model covariate (https://jinmiaochenlab.github.io/scTM/notebooks/stamp/example3/). If multiple samples are provided (and their distinction is included as a covariate), are the spatial graphs from each spatial sample still considered in the model?

[Chengwei94]

@katimbach

The covariate term is used to correct for batch effects, so in the SMI data, there is no batch since there is only 1 slice. The model is agnostic to the graph built, so if you want build the separate graph for each batch, sq.gr.spatial_neighbors(adata, library_key="data") does that. The library key builts disjoint graphs for each batch there.

[katimbach]

@Chengwei94 Thanks so much for your fast reply! Noted that I can merge samples and build the graphs after using the key.

So, I suppose in the mouse example the "data" is the obs with the slice info, as is the "library_id" in the multi-sample example (https://jinmiaochenlab.github.io/scTM/notebooks/stamp/example6/)? I was just getting confused by the "data" naming aspect (thinking it was another layer or something), but I think I understand now this is just an arbitrary name. I suppose then the neighbors for the multi-slice would've been previously built by sq.gr.spatial_neighbors(adata, library_key="library_id") ?

[Chengwei94]

@katimbach

Yep, you are right on that

[SiaGuo]

@SiaGuo,

You can just use the usual scanpy preprocess. However, the input into the algorithm is counts data, so rmb to save the counts data in the layer = "counts". Then for multiple samples, you can set the categorical_covariate_keys= [your_batch]. We will update our docs soon to make it more clear.

Hi, I've encountered with a normalized dataset for the data availability. I wonder if STAMP works on normalized ST data (the technique with fair low resolution). Thanks!