Dear scRNAseq colleagues,
My team has developed a technical framework that takes UMI matrix (dge type or 10x genomics type) as input to predict the cell type by the expression profiles. Using this tool, you don’t need any guidance from bioinformaticians to understand the cellular composition of your sample. All you need to do is upload and click
The model can characterize 42 cell types (with sufficient training data) based on the scRNAseq datasets from the public domain. We have incorporated 200k labeled expression profiles for training. Each type has >500 training cells. Nonetheless, the cell labeling was done by other researchers. We did identify and correct some of the original erroneous labels. So far, we feel comfortable to present this beta version. We had the preliminary test and the result makes sense to us, though it needs more experts’ eyes to look. The technical details will be submitted to a journal in a couple of weeks.
By using both username and password as ‘demo’, you may be able to tour the functionality of the tool. We included a PBMC dataset and a tumor dataset. You can also load your own data by register your own account. The UI was built on shiny-server and the visualization is implemented on Seurat.
I would encourage you guys to test this tool and send the valuable feedback. The collaboration is welcome if you find this tool useful.
As we are in a very low budget to maintain the server, please be advised not to load unnecessary datasets to avoid heavy consumption of the resources.
If anyone can find this tool useful and help me to broadcast to other related forums, that would be also be appreciated. Feedbacks and comments should be sent to email@example.com
-Wei Lin, PhD