GA4GH Product in Action: Individual-Centric Genomics Platform JPN

21 Jul 2026

The GA4GH Implementation Forum’s (GIF) Products in Action series showcases real-world implementations of GA4GH standards or policy frameworks. In this Product in Action blog post, Senkei Umehara of GENEX, Inc. discusses a genomics platform powered by artificial intelligence, which adopts GA4GH-compliant knowledge representation specifications to establish ethical transparency and scientific rigor, and to make AI-assisted interpretation verifiable.

By Senkei Umehara, GENEX, Inc.

Within our implementation of the Individual-Centric Genomics Platform JPN, our aim is to develop a genomics platform powered by artificial intelligence (AI) to help users own, leverage, and act on any data about themselves.

This project adopts GA4GH-compliant genomic knowledge representation specifications, including the Variation Representation Specification (VRS), Categorical Variation Representation Specification (Cat-VRS), and the Variant Annotation Specification (VA-Spec). These implementations support the move beyond conventional, often opaque flat data to establish an AI-ready genomic knowledge base with ethical transparency and scientific rigor. 

This structured approach provides a versatile foundation for applying data to cross-border analysis and knowledge federation, while accommodating local requirements like Japanese policy and language. We aim to share this agile implementation experience from Japan as a practical demonstration for building global knowledge utilities and seek feedback from the community.

Trust and transparency in knowledge are becoming increasingly vital as generative AI emerges. In promoting individual genomic data use, we support individual autonomous health management by building a platform that embodies the principle “Trustworthy AI starts with structured data.” Specifically, we designed and implemented a scalable knowledge structure based on VRS, Cat-VRS, and VA-Spec to clarify the basis for interpretation and reasoning, which is essential for targeting agentic AIs. The reason is simple: the danger is not that a model refuses to answer, but that it answers fluently and wrongly, in language too plausible to invite a second look. Those of us most at home with these tools are the least likely to notice, and we are no exception.

To maximise our limited resources we adopted a strategy of positioning the GA4GH Genomic Knowledge Standards (GKS) product suite as the core source of our competitive advantage. Starting in 2024, we spent one year deeply engaging with the GA4GH community to learn and apply the specifications. Notably, we leveraged the inherent flexibility of VA-Spec to customise its scope to address the specific requirements mandated by the Japanese rare disease policy context. Internally, we secured organisational support — our primary partnership — by sharing product development achievements with management to align with corporate strategy.

The most significant outcome is the improvement in “AI readiness” of our entrusted datasets. Compared to flat data, VA-Spec compliant structured data significantly simplifies the AI-driven extraction of actionable recommendations and clear logical rationales, thereby increasing the trustworthiness of clinical interpretation. The basis of each recommendation can be traced rather than taken on faith. We have already begun piloting this foundation for diagnostic decision support for rare disease, such as mitochondrial diseases. 

Through this experience, we also acquired the expertise for building a knowledge localisation and translation layer. It maintains international standard compatibility while meeting local requirements such as language, regulations, and region-specific disease classifications (“Nan-Byo” in Japanese). We believe this know-how offers a universally applicable solution for newly entering countries and for federated data analysis integrating heterogeneous data sources that utilise GA4GH standards. 

There were several lessons learned and recognised challenges. Active participation in regular remote meetings and exploratory adoption of latest examples greatly aided our implementation. Crucially, AI-assisted transcription and summarisation of meeting recordings and minutes provided immense help to non-native English speakers in understanding and tracking progress. The time difference between Asia and the rest of the world made real-time discussion a bit difficult. However, we found that establishing relationships through face-to-face meetings dramatically smoothed communication, covering a significant portion of the challenges posed by asynchronous communication. Conversely, fully engaging with the community (e.g. hackathon participation) remains too high a hurdle for a startup’s resources. 

A follow-up insight is that testing VA-Spec in a local context confirmed that the standard possesses the flexibility to be extended not only to global challenges but also to local regulatory and cultural requirements.

We welcome collaboration and engagement from all parties interested in building an AI-ready knowledge base using GA4GH standards. To that end, you might want to watch for the emerging GKS Starter Kit for concrete patterns to build on: it is just beginning to collect real-world vignettes of these standards in use, with more implementation examples, including ours, on the way. Through our local implementation experience, our aim is to contribute to the realisation of global data federation and knowledge equity (FAIR principles) while also seeking to sustain the commercial viability necessary for our ongoing operations. We seek to collaboratively realise a future where the public and academic ecosystem and continuous corporate activity are mutually complementary. 

—

Further Reading

Individual-Centric Genomics Platform JPN: A 2025 Implementation Progress Report