This package allows deploying a set of pre-defined foundation models for various tasks in R. It provides an interface to easily access and utilize these models using:
basilisk(https://bioconductor.org/packages/release/bioc/html/basilisk.html)reticulate(https://rstudio.github.io/reticulate/)
and for tasks such as generating embeddings of spatial or imaging datasets.
The package is designed to be user-friendly and efficient, making it accessible to both beginners and experienced users in the field of machine learning and deep learning.
Both basilisk and reticulate are used to manage Python dependencies and provide a seamless interface between R and Python. However, for foundation models that are deployed with reticulate only (e.g. KRONOS), example scripts for building and running environments can be found under inst/scripts/venv_builds.
Currently there are environments for the following models:
- scGPT (https://github.com/bowang-lab/scGPT)
- Novae (https://github.com/prism-oncology/novae)
- KRONOS (https://github.com/mahmoodlab/KRONOS)
- Nicheformer (https://github.com/theislab/nicheformer)