Efficient and Scalable Single-Cell RNA-seq Analysis Pipeline Exploiting Bayesian Non-Negative Matrix Factorization
To enable the analysis of single-cell expression data of the Human Cell Atlas by developing an efficient and scalable single-cell RNA-seq analysis pipeline powered by Bayesian non-negative matrix factorization (scRNAlyzer).
Results & Resources
The Getz group focused on scaling computational workflows for calling cell types and states from single-cell data. The core of the project was to extend current work from CPU to high performance GPU workflows, including using major machine learning packages such as tensorflow and pytorch. This was achieved with a large improvement in speed, as demonstrated in their preprint. The code for the GPU-based QTL mapper and the signature analyzer are both openly available.