DOME Registry Roadmap 2025-2026

DOME Registry Roadmap 2025-2026 1. Document Information Version ● 0.1 (draft) - 20251008 ● 0.2 (post-SAB draft) - 20251028 ● 1.0 (post-SAB feedback amended) - 20251215 Authors: ● DOME Registry Operations & Content: Gavin Farrell (UNIPD) & Omar Attafi (UNIPD) ● DOME Registry Leads: Fotis Psomopoulos (CERTH) & Silvio Tosatto (UNIPD) ● SAB: Chris Hunter (Publishing), Daniel Garijo (UPM), & Sirarat Sarntivijai (Boehringer Ingelheim) Purpose: This document defines the DOME Registry roadmap for the October 2025October 2026 period. It details aspects such as DOME Registry stakeholders, gaps, technical improvements, future implementations, journal engagements and more. It will set the one year plan to improve the service provisioning of DOME Registry with clear and achievable goals aligned to the 2026 vision. Table of Contents 1. Document Information 2. Stakeholders & Needs - Opportunities for DOME Registry 3. High Level Goals 4. Roadmap of Activities & Developments A. Technical B. Community C. Governance Roadmap Timeline 5. Potential Risks & Dependencies 6. Progress Review, Prioritisation & Monitoring 1 DOME Ecosystem The DOME ecosystem is a cohesive set of complementary research infrastructure services focused on AI/ML. They are co-developed through the ELIXIR Network and operated by UNIPD (IT) with support from CERTH (GR). ● DOME Registry: The flagship AI/ML resource that functions as a centralised FAIR resource for creating and sharing structured disclosures of AI/ML methods using the DOME Recommendations standard. It is primarily operated by UNIPD, with strategic co-design, resilience, and mirroring support from the CERTH. ● DOME Recommendations: A simple front end website that provides an overview of the DOME best practices for creating and publishing papers with high-quality AI asset transparency supporting reuse, reproducibility and trustworthiness. This standard is the basis for the DOME Registry. ● OSAI (Open and Sustainable AI): A framework of principles and practical resources designed to help implement AI/ML methods transparency, reproducibility, and environmental sustainability. It provides complementary and grounded practical support complementary to DOME. It is the grounded ‘how to’ in support of DOME. 2. Stakeholders & Needs - Opportunities for DOME Registry Stakeholders & Workflows Several DOME Registry stakeholders exist. Below are their relation to the DOME Registry and their usage workflows for using the service: Publishers: DOME Registry in AI publications Publishers are possible adopters of the DOME Registry into their publication process to enforce standards for transparency and reproducibility in AI. The DOME Registry has a specific process for onboarding and reference usage workflow from GigaScience journal pilot adopters. 1. Adoption workflow 2. Reference use workflow (GigaScience) Reviewers: DOME Registry in support of their reviews While not directly designed for lone reviewer use, the DOME Registry could be used by reviewers of manuscripts. The DOME standard can act as a checklist for their reference even if not creating a direct entry. 1. No established workflow 2 Community curators: DOME Registry for personal annotation These are authors or event organisers (e.g., from the CAID benchmarking competition) who use the DOME Registry to formally document and share their own AI methods. 1. No established workflow - but details as follows: Benchmarking event example 1. Benchmarking event instructs use of DOME Registry for event entries. 2. Benchmarking event entrants use the DOME Registry and annotate their entered paper into the event in advance for use at the event or after the event in support of publishing. 3. After submission this is released by the DOME Registry staff. Lone entry example 1. A user wants to annotate one of their own papers in the DOME Registry to adhere to DOME best practices and create a transparent disclosure of their methods. 2. The user annotates their own paper in the DOME Registry. 3. After submission this is released by the DOME Registry staff. Community curators: DOME Registry for third party annotation These are proactive community members who annotate existing papers to enrich the scientific record and gain formal credit for their curation efforts. 1. No established workflow - but details as follows: 2. A user wants to annotate AI/ML papers in the DOME Registry to receive credit via APICURON on their ORCID by supporting DOME best practices and creating a transparent disclosure of AI/ML methods. 3. The user annotates a chosen third party paper in the DOME Registry. 4. After submission this is released by the DOME Registry staff. DOME Registry staff curators: DOME Registry for third party annotation Staff curators are internal team members responsible for maintaining the release, integrity, quality, and strategic growth of the DOME Registry's content. 1. No established workflow - but details as follows: 2. A DOME Registry staff curator wants to annotate AI/ML papers in the DOME Registry in support of a meta-analysis on AI/ML papers on a specific set of models. E.g. industry models (LLMs). This can be used for a cohesive publication and would be the primary reason for DOME Registry staff curation. 3 3. The DOME Registry curation staff annotate a chosen set of third party papers in the DOME Registry. 4. After submission this is released by the DOME Registry staff. General users: DOME Registry for finding, understanding and reusing/reproducing methods General users are the primary consumers of the Registry researchers, students, and developers seeking to find, understand, and apply existing AI solutions. 1. No established workflow - but details as follows: 2. A user leverages the DOME Registry search functionality to identify methods of interest. 3. The user learns from this, discovers possible assets for reuse and possible reproduction. 4. The user may bulk download or use the API for information in meta-analyses in advanced cases. (Note: content coverage is currently too low for this to be very effective or a true use case) Policy Makers: DOME Registry as a mandate for Open Science Policy makers are institutional or governmental bodies that view the DOME Registry & related guidelines as a vital infrastructure for implementing and monitoring Open Science mandates for AI research methods. 1. No established workflow - but possible uses: 2. Policy makers discover DOME (e.g. at a policy event). 3. Policy makers can introduce DOME Registry use as part of AI/ML related outputs. Funders: DOME Registry as a requirement (like DMPs) Funders are national/international organisations that can require grant recipients to use the DOME Registry to ensure project AI methods and related outputs are managed responsibly for maximum return on investment. 1. No established workflow - but possible uses: 2. A funder is made aware of the DOME Registry and its use. 3. The funder introduces it as a requirement in grant allocations to be completed for AI/ML projects. 4. The funder could introduce it as a mention in a grant proposal text (e.g. EC have done so for a HORIZON EUrope 2025 grant call) Existing Gaps & Needs 1. Metadata 4 a. Granular template: current DOME template entry fields have several Qs under each entry which must be reduced to ensure proper answering. b. Free text: must be reduced to ensure more cross comparable searching and understandability. (e.g. fixed terms for model names vs variable free text) c. Fixed vocabulary: to support free text reduction and concepts in use for the resource. i. e.g. ontologies like AIO/MLSO/MLO/EDAM - pose suitable options. d. Machine readable metadata: improvements needed by using emerging standards (Croissant/FAIR4ML/+) 2. Publishers a. Single publisher: poses a content risk as GigaScience & GigaDB are the only adopters. New journals must be integrated to use DOME Registry. b. Ease of use: must be accelerated for adoption - manual curation is burdensome and leads to variable quality. 3. Community a. Interactions: regular and clearer interaction to the user base is needed. b. Formal support: can be improved via formalised helpdesk system. c. Training & events: 1-2 workshops a year needed to keep DOME relevant and users adopting it with reusable materials. 4. Interoperability a. AI Assets & Deposition: can be better deposited and cross linked to DOME Registry entries. (e.g. via partnership with BioAI repo of EBI). b. Publication discovery: EPMC linking to DOME entries and vice versa will yield best impact for bidirectional publication and DOME entry discovery and use. 5. Versioning and amendments a. Mechanism gap: DOME Registry entries should have versioning for updates and finer curation updates e.g. if LLM fine curation after initial submission or professional curator updates. b. Audit trail & provenance: in addition to version control there is a need to track who makes amendments, when and why. c. Authorship claim: to ensure attribution is clear and who has annotated e.g. their own publication’s model or another’s via community curation. 5 6. Scaling curation - a barrier to adoption a. Manual curation: is needed for DOME Registry entries LLMs pose the best way to accelerate this process. b. Overhead barrier: this acts as a barrier to both user and publisher adoption as manual curation is time consuming, reducing content scalability. AI Assets & Relevant Resources for Interoperability Many possible AI assets exist that DOME can either host or connect with. These are as follows: AI Publications & Supplementary Materials AI methods publications are the succinct write up and glue that holds together the public sharing of an AI method. These vary immensely between author and journals alike as no common format is established with the sharing and inclusion of AI assets such as datasets and code often varying greatly also. These are the primary sources that DOME Registry entries aim to supplement for their entries. Sources: Europe PMC - PDFs & XML AI Datasets AI methods are reliant on datasets. These are usually amalgamations of various data sources to create a reliable model that can generalise predictions based on the underlying data patterns. Their source and formats are diverse and stored in many different ways from within GitHub repos, on Zenodo or even dedicated platforms such as BioAI Repo & HuggingFace. Sources: BioAI Repo - diverse formats | Kaggle - diverse formats| HuggingFace - diverse formats| Zenodo - agnostic repository of diverse file types | + AI Model Code AI methods are reliant on code to collect the dataset, train the model and perform evaluations. This is the key underpinning for AI methods. A final model can then also be shared as the key AI assets output. These are usually shared via code repositories and in extremely diverse structures. However, given their complexity, scale and metadata they should be shared in a more structured manner such as dedicated registries. Sources: GitHub - code repository of diverse file types | GitLab - code repository of diverse file types | Zenodo - agnostic repository of diverse file types AI Metadata files 6 These are the data about the AI assets and primarily of interest are those related to the AI-datasets and models. Several emerging formats exist such as Croissant for datasets and FAIR4ML for models in varying stages of maturity. Sources: Kaggle - model cards, croissant and dataset cards| HuggingFace - model cards and croissant For more detail on many of these assets refer to the OSAI AI ecosystem. 3. High Level Goals Several high level goals exist for DOME Registry to achieve in this roadmap within the next year. These are as follows: 3.1. Enhance Platform Capabilities and User Experience ● Focus on improving the core infrastructure of the DOME Registry to make it more robust, scalable, and intuitive. This includes streamlining the submission process for contributors and enhancing the interoperability for users looking for AI assets. 3.2. Drive Growth in High-Quality, FAIR AI/ML Asset Submissions ● Increase the volume and diversity of metadata hosted on AI models and tools cataloged in the registry. This goal emphasises not just quantity, but quality, by actively promoting the use of DOME Recommendations and OSAI principles to ensure all new assets are highly reusable and reproducible. 3.3. Strengthen Community Engagement and Adoption ● Establish the DOME ecosystem as the central hub for the Open and Sustainable AI community. This involves targeted outreach, developing comprehensive training materials, and fostering collaborations with research institutions, publishers, and other ELIXIR platforms to embed DOME in standard research workflows. 3.4. Expand Integration Across the Wider AI Research Infrastructure Ecosystem ● Deepen the interoperability of DOME with key external platforms and resources. This includes connecting data exchange across assets via APIs for programmatic access and integrations with data repositories, computational environments, and publishing platforms to create a seamless experience for researchers using AI in their work. 7 4. Roadmap of Activities & Developments This section outlines the specific deliverables and activities planned for the DOME Registry, grouped into three core task groups: A. Technical Framework, B. Community of Support, and C. Governance. These items represent the activities and development roadmap for the 2025-2026 period. High-priority items have been explicitly marked as they represent the most critical tasks for enhancing the platform's stability, scalability, and adoption. While this roadmap aims to achieve a critical level of DOME acceptance of common practice in a larger community, it retains multiple implementations to be conducted in gradual phases. The roadmap provides a vision reference of the foreseeable goal to be implemented based on resource availability. It is understood that not all items may be fully achieved within the 2026 timeframe; therefore, this plan also acts as a crucial resourcing flag for funders and partners, highlighting the support required to best achieve these goals, either within this cycle or in the next 2027 DOME Registry roadmap. A. Technical Framework ID Category Activity Priority Timeline Resourcing (PM) A.1 Technical Entry Versioning & Clear Labelling Medium Q3/4 2026 ~3 A.2 Technical Core Stability - Node JS & Containerisation High Q1 2026 ~2 (Core) ~3 (Mirror) A.3 Technical Free Text Reduction & Metadata Standards High Q1/2 2026 ~6 A.4 Technical Scalable Curation - ML/LLM Support High Q1 2026 ~12 A.5 Technical Resource Interoperability - EPMC & BioAI High Q1/2 2026 ~2-3 8 A.6 Technical Resource Interoperability - AI4EOSC & MLEntory Low Q4 2026 ~3-6 A.7 Technical Handling Wider AI Assets Low Q4 2026 6-12 A.8 Technical Weighted Scoring Adjustment Medium Q3 2026 3 A.9 Technical Paper & Journal Metadata Ingestion (Crossref) Medium Q3 2026 3 A.10 Technical Ecosystem Navigation - Wizard & Registry Links High Q2 2026 2 A.1. Entry Versioning & Clear Labelling ● Priority: medium ○ Q3/4 - 2026 ● Description: Implement a publicly facing DOME Registry entry versioning system to more publicly track and display different versions of a submitted registry entry. This allows users and curators to view the history of changes and updates to an entry. There should also be functionality to re-edit and resubmit in conjunction with DOME Wizard. This is especially important for the LLM generated content for user or curator refinements vs base output. ● Why important: a critical gap identified is a mechanism for publicly facing versioning on the DOME Registry UI (Section 2). This is crucial for provenance, traceability and iterations of the content, allowing entries to be corrected or updated (e.g., after LLM generation or by professional curator intervention) rather than remaining static and potentially incorrect. The DOME Wizard provides versioning support but not translated to the front end at this time. ● Impact: Increases the long term trustworthiness and quality of registry entries. This supports High Level Goal 3.2 (Drive Growth in High-Quality... Submissions) by creating a mechanism for quality refinement. ● Resourcing (person months): 9 Pistoia Alliance. This includes assessing the need for a reusable DOME Registry internal instance to support industry stakeholders' internal transparency and local use. This could also provide a longer term pathway for eventual public release of non-trade-secret DOME Registry entries to the public platform in support of open science. ● Why important: Industry is a major producer and consumer of AI/ML methods. Engaging them directly addresses the stated goal of exploring DOME Registry for industry. It broadens DOME's applicability beyond academia, making it a more universal standard. This will also build co-development support by having more industry input to the DOME Registry development plans. E.g. info on metapredictors for pharma are a major point of importance in support of transparency of AI/ML methods. ● Impact: This opens a significant new adoption pathway and potential sustainability model (e.g. via local instance set up consultancy) (Goal 3.3). A reusable instance model could support internal industry transparency, while also enriching the public DOME Registry with high-quality, impactful methods from industrial publications, increasing its value and content base (Goal 3.2). ● Resourcing (person months): ○ 12-24 PMs C. Governance C.1 Governance SAB Meetings High Q4 2026 0.05 C.2 Governance Publisher Input Meetings High Q2 2026 3 C.1. SAB Meetings ● Priority: high ○ Q4 - 2026 ● Description: Ensure annual SAB meetings take place and feed the roadmap maintenance. ● Why important: The DOME Registry SAB (Scientific Advisory Board) provide essential high-level strategic guidance and expert external validation for the project. 16 ● Impact: Ensures the project remains on track, aligned with its high-level goals, and responsive to the needs of the wider scientific community. It provides critical oversight. ● Resourcing (person months): ○ .05PM C.2. Publisher Input Meetings ● Priority: high ○ Q2 - 2026 ● Description: Ensure communication channels, input/dialogue meetings and other points of support in place to ensure publisher input where using the DOME Registry. ● Why important: Given that "Adoption Risk (High)" (Section 5) is the main threat, keeping current and potential publishers (Stakeholders) happy and engaged is a top priority. ● Impact: This directly mitigates the "Adoption Risk." It builds strong relationships with key partners, supporting Goal 3.3 (Strengthen... Adoption) and ensuring the registry meets the needs of its most important content source. ● Resourcing (person months): ○ 3PM Roadmap Timeline DOME Registry guiding development timeline - using estimates for resourcing, completion quarters and prioritisation. This is subject to updates and amendments. 5. Potential Risks & Dependencies This section outlines the potential risks, and key dependencies that will influence the successful execution of the 2025-2026 roadmap. Risks ● Adoption Risk (High): The Registry's value is proportional to its content. Failure to onboard new publishers beyond GigaScience could lead to content stagnation, limiting the registry's utility and impact for general users. 17 ● Technical Risk (High): The success of scaling curation via LLMs is uncertain. There is a risk of generating inaccurate or low-quality metadata, which could damage the registry's reputation and require significant manual correction, defeating the purpose of automation. ● Resource Risk (Medium): The roadmap is ambitious and dependent on a small, specialised team. The loss of key personnel or insufficient funding to support the required development effort could cause significant delays or force a reduction in scope. ● Community Risk (Medium): If the curation process, even when accelerated by AI, is still perceived as too burdensome by authors and researchers, voluntary adoption will remain low. The registry may fail to achieve the critical mass of community-contributed content needed for long-term relevance. ● Scope Creep (Low): The large number of interoperability goals and new features could spread development resources too thin, risking delays across all deliverables. Careful prioritisation will be essential. Dependencies External Collaborations: ● Publishers: The growth of the registry is highly dependent on the willingness of new journals to integrate the DOME Registry into their editorial and production workflows. ● Interoperability Partners: The successful implementation of bidirectional linking relies entirely on the technical cooperation, API availability, and sustained commitment from external resources like EPMC and BioAI Repo. Technology & Standards: ● Emerging Standards: The timeline for implementing FAIR4ML and Croissant is dependent on the maturation and stable release of these community standards. ● Third-Party APIs: The automatic ingestion of publication metadata is dependent on the continued stability and accessibility of the Crossref API. 6. Progress Review, Prioritisation & Monitoring Progress will be monitored and reviewed by: ● DOME SAB during annual meetings. ● By Silvio Tosatto at the biweekly meetings with Gavin & Omar at UNIPD. 18 ● By Fotis Psomopoolos & other ELIXIR Europe parties during the AI Ecosystem FG call or adhoc calls as needed. ● Granular tracking in the GTD, GItHub and other relevant development tracking systems for technical tasks. 19