Develop an LLM-based framework for translating KNIME workflows into Galaxy Imaging workflows

Develop an LLM-based framework for translating KNIME workflows into Galaxy Imaging workflows Khanh Xuan Nguyen1,2, Matthias Bernt3, Thomas Schnicke1, Robert Haase2, Riccardo Massei1 NFDI4BIOIMAGE is funded by the Deutsche Forschungsgemeinsschaft (DFG, German Research Foundation) under the National Research Data Infrastructure, grant number NFDI 46/1, project number 501864659 1, 2 Helmholtz Center for Environmental Research (UFZ) - Research Data Management Team - Leipzig, Germany 2 Center for Scalable Data Analytics and Artificial Intelligence (ScaDS.AI) - Leipzig, Germany 3 Helmholtz Center for Environmental Research (UFZ) - Department of Computational Biology and Chemistry - Leipzig, Germany KNIME and Galaxy are leading workflow platforms widely used for imaging, data analysis, and automation. However, differences in their architectures make translating imaging workflows between them difficult and timeconsuming. The lack of automated conversion workflow reuse. Enabling automatic conversion of KNIME imaging workflows into Galaxy could enhance accessibility, streamline image analysis with further Open Source Tools (i.e. OMERO, INTOB), and accelerate imaging-based scientific discovery. Interoperability between KNIME and Galaxy Our goal is to develop generative AI strategies for automated KNIME-toGalaxy workflow translation using large language models (LLM). By learning from test workflows, the system will adapt to platform changes and improve accuracy over time. This approach: ●Reduces manual translation time and required expertise ●Ensures accurate alignment with target platform standards ●Enhances efficiency and scalability ●Continuously improves through machine learning AI to bridge the gap KNIME and Galaxy Fig.1 Current AI translation workflow integrated in a notebook Preliminary results Future Outlook Wanna contribute? Check our GitHub repo! KNIME workflow Galaxy Workflow PROMPTING Powerful AI, can you please translate this workflow to Galaxy? ●We developed a workflow in a reusable Jupyter Notebook (Fig.1) ●Our preliminary results demonstrate that the KNIME-to-Galaxy translation workflow functions as intended. ●The implemented pipeline successfully extracts KNIME node configurations, processes them via LLM-based mapping, and generates valid Galaxy workflow structures. ●At this stage, the approach works reliably for simpler workflows with straightforward node connections and limited nesting. Future work will focus on integrating retrieval-augmented generation (RAG) to improve tool and different ecosystems integration within UFZ and beyond. Further goals are to increase the node mapping, extending the translation to complex workflows with nested and conditional structures, and adding automatic validation to ensure correctness and reproducibility. Large Language Model