Only processed data is provided

HIPC scRNA-seq Annotation Benchmark Challenge
1

Hi there,

I noticed that the provided data has been processed. Some h5ad/rds didn’t include raw counts, and genes were filtered. In some data, only about 1200 genes were retained. I believe at least raw counts of all genes are needed for the analysis. Could you check the data?

Best,

Yoshi

2

Hi Yoshi,

Thank you for reaching out! For some datasets, the original authors provided raw data which we processed in-house, while for others, only pre-processed data was made publicly available. In the latter case, we are unfortunately limited to what was available to us.

Update (April 17): We internally are going over each dataset to check for which dataset only processed data was available and for which datasets both raw+processed data available. We will update you this next week.

3

Thanks. Honestly, without the complete raw count data, we can’t move forward with the challenge…

4

This is also a problem for us: we rely on raw count data for our labeling. Also, is there a way to know how the data has been processed in the h5ad objects?

5

Thanks so much for your feedback, and sorry for the slow reply on our end, as we wanted to take the time to properly gather and implement everything from Webinar 1 before getting back to you.

  1. We have made raw counts available for all studies except infection_study_07, for which we unfortunately couldn’t arrange raw count data. Teams that had been assigned infection_study_07 have been given a different study to work with, and we have opened up infection_study_07 to everyone as a bonus. If you’re able to take a look at it, fantastic, but no pressure, it’s entirely optional.

  2. To keep things comparable and reproducible across methods, we have applied a standardized QC pipeline to all datasets in the benchmark. Specifically, we applied following steps: i) removed cells with fewer than 400 detected genes to exclude low-quality or empty droplets; ii) removed cells with total UMI counts ≤200 to exclude barcodes with insufficient sequencing depth; iii) identified mitochondrial genes by the “MT-” prefix and calculated per-cell mitochondrial read percentages; and iv) filtered out cells with ≥15% mitochondrial reads, since elevated mitochondrial content usually points to damaged or dying cells that have lost cytoplasmic RNA.