Work package 2. Development of the DPOs Translator and Recommender

🎯 Objectivs

WP2 developes an AI recomendation system that take advantage of the WP1 DPOs data lake to suppoert cross-domain omics research. It translate the search quieries into domain specific isntructions across OMICs fields. Allowing researchers to optain the most relevent DPOs workflow for a single domain or compine different domains to create a containarised DPO. Benchmarking and expert validation ensure recommendations are accurate, relevant, and reproducible.

🚧 Tasks

Task 2.1 – Cross-domain DPO Translator

2.1.a – Design a linguistic framework to annotate data objects and tools across different omics domains (genomics, transcriptomics, proteomics, metabolomics). 2.1.b – Train a natural language processing (NLP) model to translate scientific questions into domain-specific instructions for DPO retrieval and workflow execution. 2.1.c – Integrate the translator with the DPO data lake to enable accurate cross-domain mapping without hallucination.

Task 2.2 – DPO Recommender

2.2.a – Develop an AI-based recommender system to suggest relevant DPOs for a given scientific query. 2.2.b – Implement mechanisms to combine or adapt DPOs from multiple domains, returning containerised workflows ready for execution. 2.2.c – Ensure the recommender operates reliably in environments with limited GPU resources, using agent-style support models grounded in the curated DPO data lake.

Task 2.3 – DPO Recommender Benchmarking

2.3.a – Conduct cross-domain pilot studies to evaluate the relevance and accuracy of DPO recommendations. 2.3.b – Collect expert assessments to verify that the recommended workflows meet domain-specific scientific needs. 2.3.c – Analyse performance metrics, document benchmarking results, and identify areas for optimisation and improvement.

🚚 Deliverables:

• D2.1 - Data Processing Obect Translator
  
• D2.2 - Data processing object Recomender
  
• D2.3 - Cross-domain benschmark/application of the Data Proecssing Object Recommeder