GeorgescuC
commented
3 years ago Hi @Sa753 ,
1) For the subcluster option, if you use one of the latest versions of infercnv (1.7.2+), the run time has been drastically reduced when using the new method (now default too) that uses Leiden clustering, so you should give it a try if you want HMM predictions. In the results you posted, there are clearly different subclones within multiple groups, but in the current state, a single set of predictions is given per group, so it is impossible to have accurate predictions for whole groups without subdividing the cell populations. At the very least you should use one of pheight/qnorm/qgamma if you find the Leiden subclustering still too slow. Step 18 is the bayesian network processing, and I expect groups that are bigger and non clonal to take longer to process, so splitting them should help speed there too.
2) For the chromosome bars, I have not encountered this issue before. Could you try running the plotting function again on the preliminary object to see if the error is reproducible? And if it is, can you try on a different machine? If it still does, I would need the object to debug the issue. There a couple cells at the top and middle of your references that show some more left over signal, so I would double check quality and cell type labeling to see if the signal is true first. For discriminating the cells within a group, you can: -use the NGCHM interactive viewer (I see you have posted an issue about it, so once that is figured out) -use the plot_per_group method to generate one plot per group, making things less compressed (you can also take advantage of the dynamic_resize option for taller heatmaps). I pushed a new commit that adds the option k_obs_group to it, allowing you to subdivide each group into the desired number. You can then look at the files generated with each plot to have the list of cells part of each subdivision.
Regards, Christophe.
andynkili
Sa753
Dear team,
I have 2 problems with running infercnv::run() function.
1-first, it takes incredible amount of time and memory. It has been running fo rover 17h and never finished for 22K cells. I am always stuck at step 18. I am running this without 'subclusters' which I don't like. The problem is even worse if I choose 'subclusters' as analysis_mode. I couldn't get beyond step 15 after 20h of running for 2500 cells. I am using 10 threads 2-Second, Is it possible to discriminate which are malignant and which are not from this graph also, why is the graph incomplete as regards to chromosome bar?(please see image)
Thanks