TL;DR
The International Mouse Phenotyping Consortium (IMPC) has rebuilt its web portal to provide more efficient and user-friendly access to its vast repository of mouse phenotyping data. This upgrade enhances the investigation of gene function and the identification of human disease models, making complex biological information more accessible to researchers.
What Problem Does This Paper Address?
The IMPC generates an enormous amount of high-dimensional phenotypic data from mouse knockout models, aiming to understand gene function and its relevance to human diseases. However, managing and accessing this ever-growing dataset (over 100 million data points from 9277 genes in the latest release) presents a significant challenge. Researchers need a way to efficiently query, visualize, and integrate this data to identify candidate disease models and gain insights into human health. The previous portal, while valuable, struggled to keep pace with the data volume and user demands for improved usability and advanced analytical tools.
Methods
The core of this paper describes the significant overhaul of the IMPC web portal. Key methodological advancements include:
- Cloud-Native Architecture: The portal was rebuilt using a cloud-native architecture. This allows for better scalability, extensibility, and easier deployment across different cloud environments, crucial for handling the massive and growing dataset.
- Microservice-Based Design: The portal now uses a component model where different sections are served by separate microservices. This improves page load times by fetching data in parallel and allows for more efficient data retrieval.
- Usability Improvements: Extensive user experience (UX) testing with both existing and new users informed design changes. This included improving color contrast, enhancing the visibility of buttons and labels, and refining navigation for easier access to information.
- Enhanced Data Visualization: New and improved graphical analysis tools were implemented. These allow users to visualize the magnitude of P-values for experimental observations, explore data dynamically with zooming and scrolling, and view detailed information by hovering over data points.
- Knowledge Graph (KG): A Neo4j KG was constructed, integrating IMPC data with other sources like the NHGRI-EBI GWAS Catalog and Open Targets. This KG uses ontologies and LLM embeddings to enable cross-cutting queries and connect mouse genes to human disease associations and shared phenotypes.
- Programmatic Access Improvements: An Apache Solr API service was enhanced, and a Python library was developed to simplify programmatic data access for users without deep technical expertise. Training materials were also updated to support this.
- Data Archiving and Accessibility: The deposition of IMPC image data into the BioImage Archive (BIA) ensures long-term availability and facilitates AI-driven image analysis.
Datasets
The primary dataset discussed is the International Mouse Phenotyping Consortium (IMPC) phenotype data.
- Data Release DR23.0 (2025): This release contains over 100 million data points from 9277 genes, identifying 113,803 significant phenotypes. This data is derived from systematic, high-throughput phenotyping of mouse knockout lines.
- Associated Data Sources: The portal integrates data from various external resources, including:
- Mouse Genome Informatics (MGI)
- Hugo Gene Nomenclature Committee (HGNC)
- Mammalian Phenotype (MP) ontology
- Human Phenotype Ontology (HPO)
- Online Mendelian Inheritance in Man (OMIM)
- Orphanet
- DECIPHER
- NHGRI-EBI GWAS Catalog
- Open Targets
- UniProt
- EMBL-EBI resources (e.g., BioImage Archive)
No new benchmarks were introduced; rather, the focus was on improving access to and analysis of existing IMPC data.
Key Insights and Ideas
- Democratizing Complex Data: The portal's redesign is a significant step towards making complex, high-dimensional biological data accessible to a broader research community, not just bioinformaticians.
- Bridging Mouse Models and Human Disease: The enhanced ability to link mouse phenotypes to human diseases (e.g., through the PhenoDigm algorithm and the disease models portal) is crucial for translational research.
- The Power of Cloud-Native and Microservices: The adoption of these architectures is essential for managing the exponential growth of biological data and ensuring the portal remains performant and sustainable.
- Usability is Paramount: The emphasis on UX testing highlights that even the most comprehensive datasets are underutilized if the interface is not intuitive and efficient.
- Knowledge Graphs as Integrators: The development of a knowledge graph demonstrates a powerful approach to connecting disparate biological datasets and enabling novel query capabilities.
Key Results
- Data Volume: The DR23.0 release contains over 100 million data points from 9277 genes, with 113,803 significant phenotypes identified.
- Disease Models: The portal facilitates the identification of candidate mouse models for human diseases, with 2938 candidate models for human one-to-one orthologs and 1435 models recapitulating at least one human disease phenotype identified in DR23.
- Portal Performance: The cloud-native architecture and microservice design significantly improved page load times and overall portal performance.
- User Feedback Integration: Usability studies led to concrete improvements in design, navigation, and data presentation, making the portal more user-friendly.
- Data Accessibility: Enhanced programmatic access via the API and Python library, along with the knowledge graph, opens new avenues for data integration and analysis.
Limitations and Future Work
- Data Volume Growth: While the cloud-native architecture is designed for scalability, the continued exponential growth of data will require ongoing infrastructure management and optimization.
- Mapping to GWAS Traits: The paper mentions ongoing efforts to map mouse phenotypes to Experimental Factor Ontology (EFO) encoded GWAS traits and integrate this into the portal. This is a significant future direction for expanding the utility of the KG.
- LLM Integration: The authors are exploring the integration of Large Language Models (LLMs) to create interactive help desks and improve data access, suggesting a future where AI plays a more prominent role in interacting with the portal.
- User Community Engagement: The authors emphasize the importance of continued user community engagement for guiding future development and ensuring existing features remain valuable.
- Completeness of Phenotyping: While extensive, the phenotyping is not exhaustive for all genes, and the paper acknowledges that the absence of assays for certain phenotypes can influence the ability of mouse models to mimic human diseases.
Summary updated