How to fix (find) number of dimensions while reducing dimensions in tSNE method?

Question

[John]

Hi, I am using Monocle to do clustering cells based on markers,

How to fix (find) number of dimensions while reducing dimensions in tSNE in Monocle (R interface)?

for example in the following command:

HSMM <- reduceDimension(HSMM, max_components = 2, num_dim = 3, norm_method = 'log',   reduction_method = 'tSNE',  verbose = T)

here, num_dim=?????

Please help, I am new to analysis, Thanks,

[Jean-Karim Heriche]

I don't know Monocle and what the parameters to its reduceDimension function are. I just want to write a note of caution because you mention both t-SNE and clustering. t-SNE is mostly useful for visualization purposes and I would question its suitability for clustering or dimensionality reduction (in the sense of further processing the data in the new low dimensionality space). The reduced dimensionality space in t-SNE has no meaning making interpretation almost impossible; in particular, it only preserves nearest-neighbours to some degree but doesn't preserve distances or densities so running any clustering algorithm relying on these after t-SNE is a bad idea. Also be aware that the output is not deterministic (i.e. several runs can produce different results). Have a look at how to use t-SNE effectively.