How would this be applied to non-Biology applications? about phate HOT 6 CLOSED

Hi Donald, just like PCA, tSNE, UMAP, and most other conditionality reduction algorithms, you can use PHATE on just about any kind of data. If you look at Supplementary Figure 3 and Supplementary Figure 10 of the manuscript (see below), you can see PHATE applied to a number of non-biological datasets.

Because you have users in a social network, I imagine your data is naturally represented as a graph. You should be able to pass the graph adjacency matrix using distance='precomputed'. Consult the docs for more information.

As for "community detection" this is a clustering task, not a dimensionality reduction task. For this, you will need to select a graph partitioning algorithm like spectral clustering or louvain.

Hope this helps!

from phate.

@dburkhardt just to clarify, given the results of a community detection algorithm, how would one represent the data with PHATE?

from phate.

I'm not sure this question makes sense. You could use the partition assignments from any clustering algorithm as a color vector when plotting, but there's no relationship between PHATE and clustering.

To color a PHATE embedding by cluster assignment, see the "Discrete color vector" examples here:

https://scprep.readthedocs.io/en/stable/examples/scatter.html

from phate.

@dburkhardt in this case, is there a pipeline for data => community detection or clustering => representation through PHATE or others. Other than that the link is useful.

from phate.

I would probably checkout our clustering notebooks from our single cell course or the scanpy clustering tutorial if you are looking for introductory clustering resources:

• https://krishnaswamylab.org/workshop
• https://scanpy-tutorials.readthedocs.io/en/latest/pbmc3k.html

from phate.

Thanks!

from phate.