Thanks for this series man. My lab recently acquired a CosMx system and this has been a tremendous help in understanding our exported data outside AtoMx. Looking forward to the next one!
You bet! This and the other sub-cellular transcriptomics I think are going to really drive some big findings these next few years. Have some surgeons already asking what can we do with the transcript locations so it's going to be exciting for sure. Really hoping to be able to use this knowledge in a career position but we shall see. Haha just fun to play with data.
Nice video, thanks. Any link to load the output from the Cosmx technology and convert to a Seurat object? How do we calculate most of the metadata metrics without having to start with the Seurat object? How are multiple samples treated ? Separate seurat objects for each sample or a merged one for all samples?
Thank you. The raw output of the technology is provided in the link in the description > Download Data > Raw Data Export. The unfortunate thing here is the size of the data; > 600 GB. To process this and create a Seurat object would require a lot of space that I am not able to get for a single run (don't have all the raw data either for my work project). I would assume that due to this constraint, working with this technology would have a Seurat object as standard. As mentioned in the video, even in the object that we worked with being 14+ GB still doesn't have the cell segmentation information. Crazy amounts of data. Also would assume that information other than the phenotyping/niche/cluster is pretty standard in the metadata - what is its immunofluorescence signal, number of probes for the cell, negative signal, etc. For multiple samples they are held within the same object. Here in the video we see a few different levels of data - Slide > Field of View > Cell > Transcripts. Certainly are able to subset to slides/FOVs but in doing this you lose a lot of information. An example of the benefits of both: I have used the individual samples with PCA and UMAP to identify malignant cells in individual fields of view where there was a mix of normal organ cells, immune cells, fibroblasts, and malignant cells; with all cells (~200k) you get more power when trying to identify phenotypes of them. To further the phenotyping problem, with only a few cells clustering would likely not separate them into a distinct group thus not allowing you to really narrow in the type but if you now combine those few cells from 10, 20, 50 samples, now you have some cells that might really stand out from the others. Tough balance between too much data (raw data here) and not enough (individual cell information) that will surely get more attention as time goes on, new methods developed, and accuracy of assays increase.
Each time I load Seurat object in rstudio it crash while loading but it is fine loading in vs code. Is it ok to use vs code or I will miss anything? Thanks
That does sound strange, huh? It's okay to use VS code for R and I think there are people that like it because of the flexibility between different programming languages. The only down side is that VS code doesn't support the R Projects (or at least last time I look - may have changed since). RStudio is pretty bad with memory (and the R language as a whole) so a really large Seurat object may be hitting the limit of RStudio while not VS code. That's my guess.
@@abdelrahmanmahany133 totally possible! Only think that will change is the .RProj part of automatically setting a working directory. Rest would work fine. Good luck!
Thanks for this series man. My lab recently acquired a CosMx system and this has been a tremendous help in understanding our exported data outside AtoMx. Looking forward to the next one!
You bet!
This and the other sub-cellular transcriptomics I think are going to really drive some big findings these next few years. Have some surgeons already asking what can we do with the transcript locations so it's going to be exciting for sure. Really hoping to be able to use this knowledge in a career position but we shall see. Haha just fun to play with data.
Nice video, thanks. Any link to load the output from the Cosmx technology and convert to a Seurat object? How do we calculate most of the metadata metrics without having to start with the Seurat object?
How are multiple samples treated ? Separate seurat objects for each sample or a merged one for all samples?
Thank you.
The raw output of the technology is provided in the link in the description > Download Data > Raw Data Export. The unfortunate thing here is the size of the data; > 600 GB. To process this and create a Seurat object would require a lot of space that I am not able to get for a single run (don't have all the raw data either for my work project). I would assume that due to this constraint, working with this technology would have a Seurat object as standard. As mentioned in the video, even in the object that we worked with being 14+ GB still doesn't have the cell segmentation information. Crazy amounts of data. Also would assume that information other than the phenotyping/niche/cluster is pretty standard in the metadata - what is its immunofluorescence signal, number of probes for the cell, negative signal, etc.
For multiple samples they are held within the same object. Here in the video we see a few different levels of data - Slide > Field of View > Cell > Transcripts. Certainly are able to subset to slides/FOVs but in doing this you lose a lot of information. An example of the benefits of both: I have used the individual samples with PCA and UMAP to identify malignant cells in individual fields of view where there was a mix of normal organ cells, immune cells, fibroblasts, and malignant cells; with all cells (~200k) you get more power when trying to identify phenotypes of them. To further the phenotyping problem, with only a few cells clustering would likely not separate them into a distinct group thus not allowing you to really narrow in the type but if you now combine those few cells from 10, 20, 50 samples, now you have some cells that might really stand out from the others.
Tough balance between too much data (raw data here) and not enough (individual cell information) that will surely get more attention as time goes on, new methods developed, and accuracy of assays increase.
Each time I load Seurat object in rstudio it crash while loading but it is fine loading in vs code. Is it ok to use vs code or I will miss anything? Thanks
That does sound strange, huh?
It's okay to use VS code for R and I think there are people that like it because of the flexibility between different programming languages. The only down side is that VS code doesn't support the R Projects (or at least last time I look - may have changed since). RStudio is pretty bad with memory (and the R language as a whole) so a really large Seurat object may be hitting the limit of RStudio while not VS code. That's my guess.
@@alexsoupir Thanks . I will try to keep up with you using vs code or I will try to increase the ram size for my device 😅
@@abdelrahmanmahany133 totally possible! Only think that will change is the .RProj part of automatically setting a working directory. Rest would work fine. Good luck!
@@alexsoupirThanks for your help
@@alexsoupirI created a swap folder and it solved the problem