Poster ID
P-34
Poster Title
Beyond Scale: How Representative Genomic Data and Federated Governance Unlock the Full Potential of Biological AI
Authors
Vinod Gauba, MD
Chairman, Count Us In Foundation
Chairman, Count Us In Foundation
Abstract
Biological foundation models are emerging as powerful tools in human genetics, yet their performance remains constrained by the representational limitations of the datasets on which they are trained. With over 95% of participants in genome-wide association studies deriving from European-ancestry populations, AI systems risk failing to generalize across diverse populations, contributing to gaps in variant interpretation, pharmacogenomics, disease biology, and therapeutic discovery.
This poster presents a framework for developing globally representative biological foundation models through federated learning, community-centred governance, and multimodal genomic and clinical data integration. The central hypothesis is that the next generation of precision medicine AI will depend as much on representative training data, trusted collaboration, and equitable participation as on advances in model architecture.
The framework is built on three principles. First, genomic and linked clinical data remain under the control of contributing institutions, enabling distributed model training without centralised transfer of raw records. Second, transparent governance and benefit-sharing mechanisms support sustained participation while respecting data sovereignty and community interests. Third, biological interpretability is strengthened through ancestry-aware variant interpretation, population reference resources, pharmacogenomic analyses, resilience biology, and downstream translational applications.
The poster outlines a federated framework for cross-jurisdictional collaboration and highlights a roadmap from population allele frequency characterisation through variant reclassification, pharmacogenomic discovery, resilience biology, and AI-enabled target prioritisation. It illustrates how globally representative datasets may uncover clinically meaningful signals, including ancestry-associated variation, pharmacogenomic patterns, protective variants, and natural human knockouts that remain underrepresented in existing reference resources.
The overarching proposition is that representative data generation, interoperable federated infrastructure, and community-centred governance are not only ethical imperatives but scientific prerequisites for building biological foundation models that produce robust, generalisable, and clinically meaningful insights across the world's populations.
This poster presents a framework for developing globally representative biological foundation models through federated learning, community-centred governance, and multimodal genomic and clinical data integration. The central hypothesis is that the next generation of precision medicine AI will depend as much on representative training data, trusted collaboration, and equitable participation as on advances in model architecture.
The framework is built on three principles. First, genomic and linked clinical data remain under the control of contributing institutions, enabling distributed model training without centralised transfer of raw records. Second, transparent governance and benefit-sharing mechanisms support sustained participation while respecting data sovereignty and community interests. Third, biological interpretability is strengthened through ancestry-aware variant interpretation, population reference resources, pharmacogenomic analyses, resilience biology, and downstream translational applications.
The poster outlines a federated framework for cross-jurisdictional collaboration and highlights a roadmap from population allele frequency characterisation through variant reclassification, pharmacogenomic discovery, resilience biology, and AI-enabled target prioritisation. It illustrates how globally representative datasets may uncover clinically meaningful signals, including ancestry-associated variation, pharmacogenomic patterns, protective variants, and natural human knockouts that remain underrepresented in existing reference resources.
The overarching proposition is that representative data generation, interoperable federated infrastructure, and community-centred governance are not only ethical imperatives but scientific prerequisites for building biological foundation models that produce robust, generalisable, and clinically meaningful insights across the world's populations.
Digital Poster