Can I merge Scissor results after performing them by subset?

[ChoiJi-Hye]

Hi,
Thank you for developing Scissor.

I ran about 130,000 cells and received a memory error message.
'Error: cannot allocate vector of size 132.0 Gb'

Then, I am thinking about running the cells in subsets and then combining the results instead of running all 130,000 cells at once.

Is it okay to combine the results after performing them by subset?
I'm worried that the results may differ from when I perform all the cells at once.

Thank you for your help.
Sincerely,
Ji-Hye Choi

[sunduanchen]

Hi You can merge several cells together to obtain a pseudo-cell or metacell, then perform Scissor on this reduced dataset. Another alternative way is to run Scissor without network regularization, which can alleviate the memory burden. Hope these help. Best, Duanchen

…

On Fri, 28 Apr 2023 at 16:26, ChoiJi-Hye ***@***.***> wrote: Hi, Thank you for developing Scissor. I ran about 130,000 cells and received a memory error message. 'Error: cannot allocate vector of size 132.0 Gb' Then, I am thinking about *running the cells in subsets and then combining the results* instead of running all 130,000 cells at once. *Is it okay to combine the results after performing them by subset?* I'm worried that the results may differ from when I perform all the cells at once. Thank you for your help. Sincerely, Ji-Hye Choi — Reply to this email directly, view it on GitHub <#50>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEYQWDCJAGEXBGLJUFI42K3XDN5MHANCNFSM6AAAAAAXO5HHGQ> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[ChoiJi-Hye]

Hi, @sunduanchen

Thank you for your response!

Could you please provide me with more information on 'Metacell' and 'Pseudo Cell'?
Does your Metacell mean the r package ‘metacell’?

I am wondering if there would be any issues if I performed subset analysis by

dividing the cells into different cell types, such as T cells and epithelial cells, and analyzing them separately,
analyzing them by sample, or
simply dividing the samples into four groups and analyzing them separately.
I noticed in another issue that you used 'Metacell' when dealing with more than 100,000 cells, but I find it a bit difficult to understand.

Also, you gave me another option, but I'm not sure where it would be possible to skip the network regularization step.
I didn't see such an option in the Scissor parameter.
Does this mean that I need to modify the source code to run it?

I'm sorry for having so many questions.
I'm very interested in this tool, but it seems a bit challenging for me to understand.

Thank you for your help.

Best regards,
Ji-Hye Choi