Skip to content
This repository has been archived by the owner on Mar 9, 2023. It is now read-only.

Latest commit

 

History

History
27 lines (15 loc) · 2.12 KB

README.md

File metadata and controls

27 lines (15 loc) · 2.12 KB

First compiled: May 3, 2017.
See the notebook.

Scanpy computationally outperforms Cell Ranger

Comparing Scanpy with the 10x Genomics Cell Ranger R kit (Zheng et al., Nat. Comms. 2017), we find that Scanpy requires 5 to 16 times less CPU time and less memory in crucial steps of the analysis. This enables analyzing 68000 cells without waiting times interactively on a regular laptop.

Design of the comparison

The analysis has been split into steps preprocessing, PCA and tSNE. Results can be compared by inspecting Scanpy and Cell Ranger and Figure 3j of Zheng et al., Nat. Comms. (2017). The benchmark runs were performed on a MacBook Pro 13-inch, Early 2015, one 2,7 GHz Intel Core i5 processor with two cores, 16 GB RAM. For rerunning the analysis, run the Scanpy and Cell Ranger notebooks. All comparisons are archived (here).

The data used for the comparison consists in 68,579 PBMC cells and is freely available here.

Speedup

We obtain a speedup in the preprocessing routines of a factor 10 to 16, in tSNE and PCA of a factor 3 to 6.

Memory

The memory measurement here only concerns the memory at the end of a step, it is not the maximum memory during a process, where R tends to allocate even a lot more. Still, here, we observe 10 times less memory used. In practice we observe an even clearer memory efficiency of Scanpy.