HPC Enhanced Genomic Selection for Climate-Resilient Crop Breeding
Organisations involved
BIOS Agrosystems SA is a Greek seed-breeding SME specialising in crop improvement and variety development. It contributes proprietary genomic, phenotypic and multi-environment field-trial datasets and validates results against commercial breeding outcomes.
bSpoke Solutions P.C. is a Greek HPC and AI specialist delivering advanced data engineering, machine learning and cloud technologies.
The challenge
Developing new crop varieties is a costly, high-risk process that typically requires 7–10 years of field testing, selection cycles and regulatory validation, with investments reaching several million Euros per variety. Climate change is increasing this risk, where unpredictable weather patterns can invalidate years of breeding work.
Evaluating thousands of genetic combinations across multiple environments creates a major business bottleneck for BIOS, limiting the speed at which improved varieties can be brought to market. The wider seed industry faces the same challenge: accelerating breeding programmes while reducing cost and improving resilience to future climates.
AI-powered genomics offers a solution by analysing genetic, field and climate data to predict future crop performance before expensive trials are conducted. However, training and validating these models requires processing millions of data points, thousands of genetic markers and numerous environmental scenarios simultaneously. Such workloads exceed conventional IT resources. High-performance computing was therefore essential to explore large model combinations, improve prediction accuracy and deliver commercially useful results within operational breeding timescales.
The solution
GENESIS is a cloud-based AI platform that enables seed companies to use advanced genomic selection without investing in specialist HPC infrastructure or data-science expertise. The platform combines genomic markers, multi-location field-trial records and climate datasets within a single machine-learning workflow to predict how candidate varieties will perform under future growing conditions.
Running on the MeluXina EuroHPC supercomputer, GENESIS can train and evaluate thousands of model configurations in parallel, dramatically reducing processing time while increasing predictive accuracy. The platform achieved prediction accuracy of up to 95.2%, meaning its recommendations closely matched real-world field results. This enabled the system to identify the most promising crop varieties before full-scale field testing, helping breeders make faster, more informed decisions while reducing the time and resources required for evaluation.
Business impacts
GENESIS delivers immediate operational value by enabling earlier identification of the most promising breeding lines, reducing the number of costly field trials required and accelerating progress towards commercial variety release.
For BIOS, during pilot evaluations across three locations, AI-generated shortlists matched actual field leaders in 90–100% of cases and identified varieties delivering yield advantages of 0.46–0.48 tonnes per hectare above site averages. These results support lower R&D costs, improved resource allocation and faster response to changing market demands.
For bSpoke Solutions, the project has created a validated AI-HPC platform ready for commercial deployment through subscription and usage-based services, opening new recurring revenue opportunities in the global crop-breeding market. More broadly, GENESIS helps smaller breeders access capabilities previously available only to large organisations, accelerating development of climate-resilient crops, supporting food security and strengthening the competitiveness of European agriculture.
Business Benefits
- 90–100% success rate in identifying field-leading varieties during independent pilot trials.
- Yield improvements of 0.46–0.48 tonnes per hectare above site averages for selected candidates.
- Up to 8× faster model training through EuroHPC resources, enabling large-scale model exploration.
- 15–20% higher prediction accuracy for European growing conditions compared with the pre-GENESIS baseline.
- Reduced breeding timelines from the traditional 7–10 years towards a faster, data-driven development process.
- Commercial SaaS/PaaS platform validated for market launch, creating new recurring revenue opportunities.