Digital Twin for Crop Growth Prediction Using Ortholog Contrast Model and Gene-Environment Interaction

Overview

This solution offers a digital twin platform for crops that predicts growth stage-based traits by combining genetic and environmental information through advanced machine learning. The approach addresses a key limitation in existing models by integrating environmental data with genetic features, enabling more accurate predictions of how crops will perform at different growth stages across diverse conditions.

The platform is particularly valuable for agricultural biotechnology companies, plant breeders, and precision agriculture firms seeking to accelerate trait selection, optimize breeding programs, and improve crop yield predictions. By leveraging functional gene scores alongside environmental variables, stakeholders can make more informed decisions about crop development and management strategies.

Technical specifications

Core methodology:

• A novel ortholog contrast model that reframes gene expression prediction as a regression task rather than classification, improving functional score estimation
• Two Tower Convolutional Neural Network (CNN) architecture for predicting gene expression variations from DNA sequences
• Integration of environmental information with genetic data through a combined machine learning approach inspired by GxENet methodology
• Transfer learning capabilities enabling adaptation across different plant species
• Correlation scores between 0.43 and 0.57 demonstrated in initial validation for predicting rank differences between species

Data sources and features:

• Ortholog pair information from databases including PlantTFDB and OrthoDB
• mRNA abundance data for species such as Z. mays and S. bicolor
• Environmental and phenotypic data inputs for growth stage-based trait prediction
• Rank difference metrics between ortholog pairs across species

Technology readiness level

The ortholog contrast model has been developed and submitted to RECOMB 2024, with initial validation demonstrating correlation scores between 0.43 and 0.57 for cross-species predictions. The foundational CNN methodology achieved 85% accuracy in classifying mRNA abundance. The project is currently in an active development phase, with planned validation including a four-month exploratory data analysis phase followed by eight months of model development focused on classifying ortholog pairs and predicting growth stage-based expression differences through gene-environment interaction. The team is actively seeking collaborations for data collection on ortholog pairs and transcriptional abundance, as well as environmental and phenotypic data.