Multi-sample experiments

[wlason]

Hi! I would like to use GASTON in a project which has multiple samples. I was wondering if you had any recommendations on how to approach the analysis - would I subset my SpatialData to a single tissue slice (I have multiple slices on each slide) and then identify domains? Or is there a way to globally identify shared domains (e.g. basal layer, immune layer, epithelial layer)? If I were to process the tissue slices individually, how would I be able to generalise the domains?

Many thanks for your help!

[uthsavc]

Hello! Thank you for your interest in GASTON.
Currently GASTON does not support multiple samples and this is something we are actively working on.

I would suggest trying two things:

1. run GASTON on the whole slide with multiple slices. In some instances, we have found GASTON is already able to learn the isodepth/domains jointly across all slices. (I would suggest trying a bigger neural network, since the data is more complex)

2. Subset your data to a single slice and run GASTON on each one separately, i.e. what you are suggesting. You could map a domain from one slice to a domain from another slice according to the overlap between the GASTON-identified spatially varying genes, i.e. the basal layers in two slices should have similar sets of continuous/discontinuous genes.

Hope this helps!

[wlason]

Thank you @uthsavc!
From your suggestions, approach 1 sounds more straightforward than 2, so I will give that a go first!
My project has ~20 tissue slices, do you have any recommendations for the number of layers and other parameters of the NN?

[uthsavc]

Oh wow, 20 slices is a lot. The default is the two neural networks have 2 hidden layers with 20 nodes each. Maybe you could start with 3-4 hidden layers with 30-50 nodes each?

Also, another alternative idea is using our new approach, GASTON-Mix (https://gaston-mix.readthedocs.io or https://github.com/raphael-group/GASTON-Mix)

[EddieLv]

@uthsavc Hi, I found that you have published MULTI-GASTON,https://github.com/raphael-group/Multi-GASTON, when could we use it?

[uthsavc]

@EddieLv Multi-GASTON is still under construction but will hopefully be ready soon!