[{"id":1044,"title":"Docker containers with embedded R packages and domain-specific tools","sha":"09ce69715f6bbf24b137859378015779","user_id":null,"state":"published","body":"R packages are collections of functions, documentations, tests, sample data, and dependency declarations, which come together to promote modularity, reproducibility, and adherence to good coding practices. Containerization using technologies such as Docker allow for building a specific computing environment with all required dependencies, configuration files, and system libraries needed to run any computing process reproducibly and at scale. Further, Docker allows computing environments to be _specified in code_, such that the instructions for building the infrastructure itself can be specified in code. Many data science and bioinformatics tasks involve running some domain-specific tools for initial analysis, followed by postprocessing using R. The `rpdd` package provides a demonstration on how to create a Docker image with an embedded R package and domain-specific tools. The `rpdd` package comes with a build script, which first builds the R package, then builds a Docker image containing that package, along with domain-specific tools and dependencies. When the container is instantiated, it runs a script that uses domain-specific tools to preprocess input data, and runs functions from the R package to postprocess the output from the first step. Code and further documentation is available at \u003chttps://github.com/stephenturner/rpdd\u003e.","created_at":"2023-02-02T14:29:21.946Z","updated_at":"2023-03-30T09:35:14.831Z","zenodo_id":7662885,"doi":"https://doi.org/10.5281/zenodo.7662885","subject":null,"tags":["docker","r packages","bioinformatics","data science","reproducible research"],"vote_count":1,"view_count":1,"score":1.1,"deleted":false,"muted":false,"tweeted":true,"attachment_file_name":null,"attachment_content_type":null,"attachment_file_size":null,"attachment_updated_at":null},{"id":1630,"title":"Converting Snakemake workflows to Nextflow with nf-core modules using a fine-tuned LLM","sha":"4724fe12f939175b4ac3c0ca6ed89a56","user_id":null,"state":"published","body":"A useful application of AI in bioinformatics is the automatic transformation of Snakemake workflows into Nextflow, leveraging nf-core modules and subworkflows where available. As bioinformatics pipelines become increasingly complex, the ability to seamlessly transition between workflow languages is essential, particularly in collaborative environments where diverse teams may prefer different tools. A fine-tuned LLaMA model or an OpenAI custom GPT  could be trained to interpret the structure and logic of a Snakemake workflow on GitHub and convert the workflow to Nextflow. This model could identify where existing nf-core modules and subworkflows can replace custom scripts, simplifying the process and ensuring adherence to best practices in pipeline design. Such a system would reduce the burden of manually translating workflows, enabling researchers to take advantage of Nextflow’s scalability and reproducibility without significant overhead.\r\n\r\nTo implement this, one could fine-tune an existing language model on a dataset of paired Snakemake and Nextflow workflows, with special emphasis on nf-core standards. The model would need to be able to understand both the structure of workflow definitions and the underlying bioinformatics tools being invoked. Additionally, the model should query nf-core’s module registry to determine where standard subworkflows can be integrated, further enhancing workflow consistency and reducing redundancy.","created_at":"2024-09-14T14:10:20.410Z","updated_at":"2024-10-17T06:00:22.213Z","zenodo_id":13942979,"doi":"https://doi.org/10.5281/zenodo.13942979","subject":null,"tags":["snakemake","nextflow","llm","llama"],"vote_count":0,"view_count":1,"score":0.1,"deleted":false,"muted":false,"tweeted":false,"attachment_file_name":null,"attachment_content_type":null,"attachment_file_size":null,"attachment_updated_at":null}]