Network-based genomic selection platform integrating deep learning and bayesian causal reasoning for precision agriculture

Overview

Genomic selection for complex agronomic traits remains difficult because these traits are polygenic, meaning they are governed by many genes each with small individual effects, and they are strongly influenced by environmental conditions. Traditional genome-wide association studies (GWAS) have struggled to capture these interactions, identifying only a limited number of genetic factors with low penetrance. This solution addresses that gap by combining deep learning with Bayesian causal reasoning to uncover coordinated gene modules, alleles, and environmental interactions underlying target traits. The result is actionable guidance for crop selection, planting strategies, and resource management such as water and pesticide use, with downstream potential for CRISPR-based genetic enhancement.

Technical specifications

Key features:

• Network-based GWAS method that identifies modules of genetic factors contributing to complex traits, rather than isolating individual genes
• Deep learning integration for detecting patterns across large-scale genotyping, gene expression, and phenotyping datasets
• Bayesian causal reasoning to model interactions between genetic factors and environmental conditions
• Multi-data fusion combining genotyping, gene expression, and in-field phenotyping data
• Cross-crop applicability with initial development on public corn datasets and extension to additional crops
• Actionable output generating testable hypotheses suitable for CRISPR-based gene editing and precision agriculture deployment

How it works: The platform ingests multi-modal agricultural data including genetic sequences, gene expression profiles, and field-level phenotypic observations. Deep learning models detect complex patterns across these datasets while Bayesian causal reasoning distinguishes true genetic-environmental interactions from noise. The system outputs identified gene modules and their associated alleles, along with the environmental conditions under which they exert influence, enabling targeted breeding and resource management decisions.

Technology readiness level

The network-based GWAS method has already been developed and demonstrated, including prior work on complex traits such as the effects of vitamin B intake on skin color. The current project focuses on extending the method to incorporate gene-environment interactions and validating it across multiple crops. Initial testing will use public corn datasets, followed by application to proprietary crop datasets with extensive genotyping, gene expression, and in-field phenotyping data. The platform is positioned for collaborative refinement and validation with industry partners, with readiness for sponsored research, co-development, and pilot trial engagements.