Hi, The link for an example in Section 3.4.2 to make a model file is not working. GitHub page is showing an error. For your reference (Example model: An example of an MCP-Counter-based solution to the coarse-grained sub-Challenge is provided here). This here link is not working. Can please provide another example of creating the model file or another link for it? Thanks

Created by Neelam Sharma neelamsharma
Hi @djo Is there any specific way to create these admixtures? It would be helpful if you can post some resource for it.
Hi @neelamsharma We are not providing the percentages of cells in each dataset. There are two types of datasets, those consisting of admixture samples and those consisting of purified samples. @djo described the purified samples. Thank you, Dominik. Purified samples include only cells of one type--e.g., CD4 T cells. We imagined you would use them as Dominik describes. The samples provided in the leaderboard and validation rounds are admixtures. Obviously, these do have associated "percentages" (or at least values that correlate with percentages) that we will use as ground truth to evaluate your methods. We are not making those ground truth values / percentages available. Best, Brian
@neelamsharma, @djo Thanks you for catching the broken link, it's been fixed. I'll let Brian respond to the rest of the questions.
Hi @neelamsharma, I am just another participant of the challenge and cannot give any official statements. By my personal understanding, the listed datasets are experimental data from different unrelated studies that happened to analyze expression profiles of certain immune cells, or in some cases extracellular vesicles, under certain conditions. Most of the samples do not constitute a known mixture of different cell types but instead expressions of isolated cell types or at least a very limited group of different cells. The challenge committee used text-mining to determine this list of studies that likely contain relevant data but it is up to us to curate the data and select samples that can be used for the approach of our choice. E.g. if you like to train your model on actual cell mixtures you could combine expression vectors of pure cells to produce in silico admixtures and use them as training data. Kind regards, Dominik
Hi, Thank you for the response. Can you please let us know where we can find the percentage of the samples/cells in each dataset? Thanks
I think this is meant: https://github.com/Sage-Bionetworks/Tumor-Deconvolution-Challenge-Workflow/tree/master/example_files/example_submissions/mcpcounter_course

Link for example for model file is not working page is loading…