Unsupervised classification of a Flow Cytometry time series data

[gmanuch]

Title
Unsupervised classification of a Flow Cytometry time series data

Summary
Hoping to identify various species as clusters in flow cytometry data and try to keep track of how these clusters change in time as the environmental variables change in a transect taken by a cruise.

Personnel
Georgy Manucharyan, University of Washington, School of Oceanography (I'm happy to lead the machine learning component of the study). Need a biologist as an additional lead!

Specific Tasks
Start by identifying clusters using a range of unsupervised learning techniques; cofienventionally, classes are identified using human-prescribe gates). Once this works, figure out how to bridge different time snapshots together to make sure the coherency/smoothness in the time evolution of the clusters. Analyze cluster characteristics in relation to environmental variables.

Reading

• https://en.wikipedia.org/wiki/Flow_cytometry

[cathmmitchell]

Cool idea - do you have a dataset in mind for this?

[gmanuch]

Yep, at the moment, someone sent me the data from the HOT297 cruise, it's about 1.5GB to get started. But I think we could find the flow cytometry data in somewhere in an open domain.

[sophieclayton]

this is cool!
I worked with some stats folks on some similar things, here's a repo for one of the projects that used changepoint detection to find distinct communities in time/space. Possibly not what you are going for

[ANetTow]

I have access to the raw SeaFlow data. The challenge I found next week was uploading to Google Drive in a timely fashion.

[ANetTow]

By "next" week I meant "last" week. Slow internet at home.

[ANetTow]

Northwest was hungry so we all rushed to get lunch before the cafeterias closed, so we missed the breakout groups. I'm moderating this project. I have access to a boatload of data and am trying to figure out the best way to share.

I know the data well but don't have expertise in machine learning or other automated tools, so I'm looking forward to hearing from you all!