Using Bioconductor On GPU Enabled Cloud Vms
Author(s): Nitesh Turaga,Vincent James Carey
Affiliation(s): Dana Farber Cancer institute
GPU-enabled virtual machines provide the ability to speed up computation for deep learning libraries such as Keras and TensorFlow. These libraries are usually written in python, but with the help of R + python interoperability packages like “reticulate” (CRAN) and “basilisk” (Bioconductor), developers have been able to leverage these deep learning capabilities in Bioconductor. In addition, RStudio provides reticulate-based ‘Keras’ and ‘Tensorflow’ packages which are very robust. Leveraging this interoperability, we’ve seen an influx of packages in Bioconductor such as VAExprs and DeepPINCS, etc. Users can now take full advantage of these packages in two ways: use the AnVIL project (www.anvilproject.org), which has GPU-enabled cloud environments, or use the docker images, which have GPU capability by using a cloud provider of choice. Finally, there is the benefit of using different python environments to use a specific python version and python library version of choice to allow backward compatibility of packages. This short talk will demonstrate both these paths to enable users to take advantage of deep learning libraries on GPU-enabled Bioconductor cloud environments.