Waddington optimal transport for single-cell analysis

Question

[Assa Yeroslaviz]

Who has experience with the Waddington optimal transport tool (wot) and might be able to help us solve a problem?

I have done all the example data set on the web site and everything worked perfectly well. Now I'm trying to use it to analyse my own data. This data is from 10x Genomics. To test the data set we sort of created a subset of all the data we have, taking into account "only" three time-points, taking into account 10482 cells with 18418 genes.

I'm having troubles creating the cell-sets needed for most of the analyses in the workflow. I have calculated the transport maps using the command

wot optimal_transport --matrix p2_matrix.transposed.txt --cell_days p2_timepoint.txt  --out tmaps --format loom --local_pca 7

Now I'm trying to calculated the cell-sets, but keep getting an empty file

the command is this

wot cells_by_gene_set --matrix p2_matrix.transposed.txt --gene_sets original_backup/p2_gene_set2.gmt --out p2_cell_sets.gmt --format gmt --quantile 0.99

but the resulted file looks like that:

cat p2_cell_sets.gmt 
Cluster1        -
Cluster2        -
Cluster3        -
...
Cluster18       -

When I create random subsets of the big gene-matrix (e.g. 100 genes, or 1000 cells) I do get some results, but almost 99% of the cells grouped into just one Cluster. Depending on the randomness of the data set I get different number of cells clustered, but not all of the cells I'm using as input.

I'm not sure if I'm missing something crucial or if there is a problem with approach I'm taking. But I would appreciate any help one can provide or ideas to solve this conundrum.

thanks

[Kevin Blighe]

You may want to modify the parameters, for example, reducing the value of quantile. Also, ensure that your input data has been QCd by filtering out outliers.

[Assa Yeroslaviz]

sorry for the late reply. do you have experiene with the wot algorithm? I have reduced the quantile to 0.01 and still got empty files. Only when reducing the number of genes (creating a random subset) i manage to get some cell. Can you explain to me, based on what parameters/factors the cells are grouped into clusters? thanks

[Kevin Blighe]

I am only vaguely familiar with the work of Waddington, to be honest. I made my comment for the fear that nobody else would respond. You may want to contact Broad Institute.