FAQ  •  Register  •  Login

FlowJo and Cytof

Forum rules
Please be as geeky as possible. Reference, reference, reference.
Also, please note that this is a mixed bag of math-gurus and mathematically challenged, so choose your words wisely :-)
<<

marodon

Participant

Posts: 3

Joined: Mon Sep 26, 2016 7:46 am

Post Wed Oct 05, 2016 2:25 pm

FlowJo and Cytof

Halo, I am using the plugins of flowjo X to analyze cytof data. At one point, we are asked to parameter the tSNE analysis with 4 different variables; iterations, perplexity, eta (learning rate) and Theta. Single values within defined range must be entered. My question is how does changing those parameters all affect the "look" of the tSNE analysis? By default, I feel that my cells are too dispersed and I would like them to have a more condensed profile. The doc page on flowjo web site is rather scarce on explanations on how to set these values. Any experience/advice? (before I run the same analysis over and over and changing one parameter at a time to see how it affects the tSNE cloud) Thanks.
<<

MikeStadnisky

Participant

Posts: 2

Joined: Tue Feb 17, 2015 10:19 pm

Post Thu Oct 06, 2016 1:48 pm

Re: FlowJo and Cytof

Great question, we ran a tSNE variable study on one file (a random downsample gate of 10,000 events in the CD4 subset) to examine the effect of tSNE variables and changed one variable at a time, and I hope this provides a bit more detail based on our experience -

iterations: In our studies, we tested iteration values of 40-200, and values below 200 led to fragmentation of known populations, so we set the default to 200. (Note that the number of iterations may be fixed in other implementations.)

perplexity: A range of values from 10 to 100 was tested. In general, lower values resulted in less coherent distributions over the length of the run; the highest values resulted in populations that were much more distant. Note that with reductions of higher numbers of parameters, it may be useful to increase perplexity to 50 or more.

eta (learning rate): Higher values of this parameter make the events move more quickly into position (possibly at the expense of accuracy?). e.g. the distribution at iteration 300 for the highest value tested of eta is about what the iteration 400 for the next highest value of eta generates and is about what the iteration 500 for the default eta value of 200 (middle) generates.

Theta: There was surprisingly little dependence on this value. The literature suggests using a value of 0.5, and this was selected as the default value for the user interface.
<<

marodon

Participant

Posts: 3

Joined: Mon Sep 26, 2016 7:46 am

Post Mon Oct 10, 2016 11:55 am

Re: FlowJo and Cytof

Thanks. Very helpful. I guess I will have to play with perplexity only (which is the day-to-day workload for researchers right? :)
<<

dtelad11

Master

Posts: 113

Joined: Mon Oct 31, 2016 6:26 pm

Post Wed Nov 02, 2016 3:34 pm

Re: FlowJo and Cytof

Here's a useful introduction to t-SNE parameters:

http://distill.pub/2016/misread-tsne/

Return to CyTOF data analysis

Who is online

Users browsing this forum: No registered users and 1 guest