Scripts and notebooks for reproducing analysis from the 2020 dropkick manuscript.

---

1. Run install.sh to install dropkick and dependencies for completing downstream analysis. It is recommended to initiate a new virtual python environment to avoid conflicts.

2. (optional) Download six replicates of human placenta data and perform CellBender remove-background by following instructions in the cellbender/ directory.

3. (optional) Build a high-background simulation from human PBMC data by following instructions in the simulation/ directory.

4. To add EmptyDrops and CellRanger v2 labels to all raw .h5ad files in data/, run emptydrops_cellranger.sh. This will deposit labels into the original .h5ad file.

5. run_and_test.sh will perform dropkick filtering on all .h5ad files in data/ and then compare outputs to CellRanger_2 and EmptyDrops labels.

6. test_3907.sh will combine the 3907 human colorectal cancer datasets and reduce dimensions, showing differences between filtering tools.

7. test_placenta.sh will combine the human placenta datasets downloaded in step 2 and reduce dimensions.

8. test_simulation.sh will reduce dimensions for the high-background simulation made in step 3 above.

9. manualfilter_example.ipynb outlines our manual filtering approach performed on inDrop sequencing samples for comparison to dropkick.

---

NOTE: This entire process may take several minutes to complete on a well-equipped machine. The emptydrops_cellranger.sh script typically completes in ~5 min per input file (.h5ad). The run_and_test.sh script may also take a couple minutes for dimension reduction, clustering, and embedding to visualize dropkick vs. EmptyDrops and CellRanger filtering.

---

dropkick Documentation

Full documentation is available at KenLauLab.github.io/dropkick.