Vireo misassigns cells to wrong donor.

[maxozo]

Hi, We have been trying to investigate on why running Vireo with a different --randSeed sometimes behaves very differently. Maybe you can help? Particularly we sporadically sometimes get dropout donors and a doubled donors as in the Figs attached. The only difference between these 2 runs is --randSeed (in one case its 1 and in another its set to 12).

From experimental procedures we know that the cell numbers should be about the same for each donor. This experiment was performed in duplicate.

Thank you! Bw, Matiss

[huangyh09]

Hi Matiss,

Thanks for sharing this. This is indeed a potential issue that the model can only guarantee a local optima solution, and different initialization may return different local optima. By default (with --nInit argument), Vireo will run 50 random initializations (based on the random seed), and usually, it will reach the (near) global optima. However, for some scenarios, e.g., many donors (>10), 50 initializations may not be enough to reach the global optima. For example, in your left panel, the "donor1" seems to be absorbed into "donor2" as a local optimum.

One simple solution is that you run the model with more initializations, e.g., --nInit 200. Then it may give a better chance to find the global optima, in most of your trials (with different random seeds).

Yuanhua

[maxozo]

Hi Yuanhua, Thank you for this. The --nInit 200 does indeed solve the issue. Just posting a screnshot of the default initiasiation on left and 200 on right.

[huangyh09]

Thanks. This looks nice. I will close the issue here but feel free to re-open (or open a new one) it if you have additional questions.