Data Simulation

[galaxywall]

Could you please tell me how I am supposed to extract one single simulated dataset from your simulation codes. I thought the gibbs components of the variable modelList are about the simulated data, but there are non-integer and negative values in the data.

I also have trouble extracting a single component of modelList. For example, modelList$gibbs does not work.

By the way, I see there are non-integer values in your shalek2014.RData. Could you explain on that?

[amcdavid]

The shalek data is log2 + 1 transformed. Yes, I believe you have correctly identified the simulated data. There's no reason that it would be integer-valued, or non-negative. The model that was proposed is a zero-inflated multivariate normal, so doesn't imply either of those properties. In applying it to the datasets mentioned in the paper, I always "conditionally centered" the non-zero observations, as is indicated in the supplement. Maybe you can explain more what you are hoping to do with the simulation code, I certainly wouldn't claim that it realistically represents all aspects of scRNAseq data, especially not UMI data. However, one could translate, exponentiate and round it, to get heavy-tailed non-negative counts, for instance.

…

On Aug 26, 2020, at 11:33 AM, galaxywall ***@***.***> wrote: Could you please tell me how I am supposed to extract one single simulated dataset from your simulation codes. I thought the gibbs components of the variable modelList are about the simulated data, but there are non-integer and negative values in the data. I also have trouble extracting a single component of modelList. For example, modelList$gibbs does not work. By the way, I see there are non-integer values in your shalek2014.RData. Could you explain on that? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#5>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AALLAHXFF2CY2GUP3V37GYLSCUTMLANCNFSM4QL62IYQ>.

[galaxywall]

Well, I am really trying to simulate data using your codes and then apply some other algorithms. I just have a few more concerns regarding your modelList and modelArg.

1, For modelList, what is that $trueFactor about?
2, For modelArgs,
i. What are those entries "dense" and "sparse" in the column "type" about?
ii. Why are there NA's in the column "P"?
iii. Do those "none"s in the column "contam" just correspond to the pairwise hurdle models?

[amcdavid]

On Aug 27, 2020, at 8:18 PM, galaxywall ***@***.***> wrote: Well, I am really trying to simulate data using your codes and then apply some other algorithms. I just have a few more concerns regarding your modelList and modelArg. 1, For modelList, what is that $trueFactor about?

I honestly can't recall but looking at the code it seems like it is an attempt to record the signs of the G/H/K interaction matrices using a set of binary flags. I don't know why I was interested in doing this. It's easy to verify that not used anywhere else in the codebase once it's set. Searching the codebase might answer other questions you have. Since I wrote this code several years ago, I'm basically in the same position you are in. https://github.com/amcdavid/HurdleNormal/search?q=trueFactor&unscoped_q=trueFactor <https://github.com/amcdavid/HurdleNormal/search?q=trueFactor&unscoped_q=trueFactor>

2, For modelArgs, i. What are those entries "dense" and "sparse" in the column "type" about?

dense = "type 2", sparse = "type 1" in supplemental table 1.

ii. Why are there NA's in the column "P"?

Not entirely sure, but I guess it's from line 23 here https://github.com/amcdavid/HurdleNormal/blob/773f623301c67642364e3cf281c6c95f9b3a5ac2/inst/aoas_simulations/simulations.R#L23 <https://github.com/amcdavid/HurdleNormal/blob/773f623301c67642364e3cf281c6c95f9b3a5ac2/inst/aoas_simulations/simulations.R#L23>. The `P` argument only applies to models simulated with simulateHurdle210, the ecoli simulations are 500 nodes so you could set P to that if you want -- it doesn't affect the simulation either way.

iii. Do those "none"s in the column "contam" just correspond to the pairwise hurdle models?

Yes, otherwise they are type 5 or 6 in supplemental table 1.

…

— You are receiving this because you commented. Reply to this email directly, view it on GitHub <#5 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AALLAHRHNVLVEPAEQYLCYFDSC3ZWPANCNFSM4QL62IYQ>.