Reproducibility PhenoGraph algorithm in histoCAT

Looking for ways to analyze your dataset? Need help with a software package?
Computational image analysis of multiplexed Imaging Mass Cytometry acquisitions
Post Reply
ElisaRasca
Posts: 8
Joined: Mon Oct 08, 2018 3:02 pm

Reproducibility PhenoGraph algorithm in histoCAT

Post by ElisaRasca » Wed Oct 31, 2018 2:32 pm

In the paper presenting histoCAT, it is stated that an internal seed is set in order to reproduce over time the results of the PhenoGraph clustering algorithm. If I'm not mistaken, this occurs at line 18 of the Run_PhenoGraph_Callback.m script on github.

However, I found that repeating exactly the same operations in different histoCAT (v 1.73) sessions does not return the same results. In particular, I notice this while plotting a heatmap of clusters x channels. Not only the clusters have different indices, but there seem to also be slight differences in the values for the different channels. I tried this a few times on different ROIs, and in all cases results were not reproducible.

How is this possible?


Elisa
DenisSchapiro
Posts: 33
Joined: Wed Nov 29, 2017 11:34 am

Re: Reproducibility PhenoGraph algorithm in histoCAT

Post by DenisSchapiro » Tue Nov 20, 2018 5:25 pm

This will be fixed in the next version of histoCAT
DenisSchapiro
Posts: 33
Joined: Wed Nov 29, 2017 11:34 am

Re: Reproducibility PhenoGraph algorithm in histoCAT

Post by DenisSchapiro » Tue Nov 20, 2018 9:45 pm

Quick update:
In certain situations, I still see the wrong behavior. I made an issue - please stay updated here:
https://github.com/BodenmillerGroup/histoCAT/issues/14
DenisSchapiro
Posts: 33
Joined: Wed Nov 29, 2017 11:34 am

Re: Reproducibility PhenoGraph algorithm in histoCAT

Post by DenisSchapiro » Tue Jan 08, 2019 7:52 pm

Update:
The new histoCAT version provides the possibility to set a seed now.
Post Reply