Inquiry Regarding Memory Requirements for CRC HD Data Deconvolution

[lllxxyyy-lxy]

Hello, and thank you for providing the Visium HD analysis tutorial—it's been incredibly helpful. I am currently analyzing CRC HD data and performing deconvolution following your demo. However, the process consistently terminates due to insufficient memory. I have made sure to set all parameters according to the Deconvolution.R script.

Could you please provide information on the hardware configuration used in your lab, particularly the memory size, to ensure that my setup is appropriate for this task?

Thank you in advance for your help!

[lllxxyyy-lxy]

In addition, my device configuration is 64G memory and 24 cores of CPU, OS: win11. When I set max_score to 4 or 1, the parallelism was reduced, but the running time was as long as 30 hours and there was no result.

[lllxxyyy-lxy]

myRCTD <- run.RCTD(myRCTD, doublet_mode = 'doublet')
fitBulk: decomposing bulk
chooseSigma: using initial Q_mat with sigma = 1
Likelihood value: 950916.436923647
Sigma value: 0.84
Likelihood value: 945802.189349382
Sigma value: 0.69
Likelihood value: 943050.320950456
Sigma value: 0.63
Likelihood value: 942519.341043391
Sigma value: 0.6
Likelihood value: 942374.320396257
Sigma value: 0.58
Likelihood value: 942321.405296159
Sigma value: 0.57
Likelihood value: 942307.982177772
Sigma value: 0.57
Error in serialize(data, node$con): Error in link writing
In addition: Warning message:
In asMethod(object) :
sparse->dense coercion: allocating vector of size 13.3 GiB

[jpromeror]

Hello! Sorry for the delay in the answer.

Looking at your output, it seems that you are using the standard spaceXR version and not the one we modified to make it more efficient for Visium HD. Please see this PR for more details. The full modified code is here: https://github.com/jpromeror/spacexr/tree/HD

Regarding the hardware, we used different severs and cloud instances with different configurations (all of them with > 64GB of RAM).

My first recommendation would be to install the version I mention above and try re-running with 8-12 cores.

Let me know if there are any issues

[jpromeror]

I will be closing this issue due to inactivity. Please feel free to reopen it if further discussion is needed.