MAGen is built as a multi-site, team-science consortium designed to develop, test, and responsibly translate ML/AI tools for genomic translational research. Our model emphasizes shared standards, independent validation, and continuous collaboration across disciplines, institutions, and data environments.
Why a team science model?
Predicting disease manifestation for individuals with pathogenic genetic variants requires more than a single dataset, method, or specialty. MAGen’s team science approach brings together:
- genomics and genetic epidemiology
- clinical informatics and learning health systems
- ML/AI method development and evaluation
- biostatistics and data engineering
- ethical, legal, and social implications (ELSI)
- implementation and dissemination expertise
By coordinating across sites, MAGen aims to produce tools that are more reproducible, generalizable, and ready for translational pathways.
How collaboration works in practice
MAGen uses consistent workflows to help teams collaborate efficiently while maintaining site independence needed for validation.
1) Shared questions, shared definitions
Teams align on target use cases (e.g., disease outcomes and cohorts), key definitions (phenotypes, exposures, covariates), and minimum reporting requirements so that results can be compared across institutions.
2) Common data expectations, local flexibility
Sites work toward harmonized data models and documentation while retaining flexibility to use local data assets (EHR-derived features, omics, environmental measures, SDOH variables, etc.) and governance processes.
3) Reproducible tool packaging
MAGen tools are designed to be portable and testable outside the originating site. Tool “packages” generally include:
- model specification and parameters
- training/inference pipelines
- documentation (inputs, outputs, limitations)
- evaluation procedures and reporting templates
4) Independent cross-site validation (a core feature)
Tools developed at one site are evaluated by other MAGen sites on distinct datasets. This structure helps determine:
- how well performance generalizes across populations and settings
- sensitivity to data differences and measurement variation
- potential bias and subgroup performance differences
- reproducibility of implementation and results
5) Iterative improvement loop
Validation findings feed back into refinement, updating feature definitions, model design, calibration approaches, and evaluation plans to improve robustness and transparency.
Responsible AI embedded throughout
MAGen integrates ELSI from the start rather than treating ethics as a final step. Responsible AI practices are woven into tool development and evaluation, including attention to:
- bias in data and algorithms
- fairness and subgroup performance assessment
- privacy and appropriate governance
- transparency, interpretability, and documentation of limitations
- trust and real-world implications of predictive tools
This integrated approach supports scientific rigor and helps reduce risk of harm or unintended inequities.
How decisions are made
MAGen uses a consortium-based approach to coordination and accountability, typically including:
- cross-site working groups (e.g., methods, data harmonization, validation, ELSI),
- shared protocols and reporting standards,
- regular scientific and operational meetings,
- milestone-driven project management supported by the Coordinating Center, and
- independent feedback from the External Scientific Panel.
What success looks like
MAGen’s team science model is designed to produce:
- ML/AI tools that can be implemented and tested beyond a single institution,
- evidence of generalizability and reproducibility across diverse datasets,
- clear documentation of appropriate use, limitations, and risks, and
- a foundation for responsible translation of AI-enabled genomic prediction into future research and, ultimately, clinical practice
Get involved
Interested in collaborating with MAGen or learning more about our cross-site validation and responsible AI approach?