Single Cell on Colab
The goal of this project is to optimize Google Colab for the purpose of performing single cell analysis.
Initially the project is integrating existing popular tools in both Python and R. The longer term goal is to build out a set of components which can be assembled into end-to-end GPU pipelines. In such a pipeline, the data would be downloaded and parked on the GPU wherein all analysis computation would happen. Apache Arrow enables such architectures, as illustrated in the following diagram from Nvidia:
For the context of single cell analysis, the types of GPU powered compenents being assembled include:
- Data parsers (e.g. nvParse for CSV, sparse coordinate matrices, etc.)
- Multi-GPU PCA
- GPGPU Linear Complexity t-SNE Optimization
- UMAP on RAPIDS (15x Speedup)vv
- cuDatashader: Datashader implemented on GPU by Rapids
Related notebooks in other repos
There are similar notebooks in separate repos:
- allensdk_on_colab.ipynb: install and debug AllenSDK on Colab
- colab_vm_config.ipynb: stats on the VM
- git_console.ipynb: manually git on Colab
- python_using_r.ipynb: Python rpy2 to R
- sleepwalk_on_colab.ipynb: tSNE & UMAP JS viz
Presentation
Seattle Cell Science Symposium 2019
The first public presentation was at Seattle Cell Science Symposium 2019
Lightening talk slide
|
Poster
|