Digital twin modelling of plants using logic programming for agricultural knowledge integration

Overview

This solution applies logic programming (LP) to build digital twin representations of plants by formalising heterogeneous biological and physiological data as logical statements and rules. The resulting knowledge base functions as a queryable knowledge graph that can integrate omics data, physiological measurements, and domain relationships into a unified, human-understandable model. By learning relational rules from this structured data, the approach uncovers interpretable connections between plant traits, genes, and functions while supporting predictive tasks relevant to crop improvement.

The technology is designed for agricultural researchers, plant breeders, and crop science organisations seeking to combine disparate data sources into coherent models that can be queried, reasoned over, and used to generate actionable insight.

Technical specifications

Core approach:

• Logic programming paradigm used to represent plant-related entities and their relationships as formal logical statements
• Knowledge corpus structured as a graph that supports querying and automated reasoning
• Integration of heterogeneous data spanning multiple biological and physiological domains
• Rule-learning capability that produces human-interpretable relational rules linking plant attributes, genes, and functions
• Binary representations of learned relations (presence or absence) that can augment standard datasets for predictive modelling

Key features:

• Demonstrated ability to learn interpretable rules such as identifying drug-gene-function pathways relevant to a target
• Improved predictive accuracy for gene expression prediction when learned representations augment standard datasets
• Designed to formalise facts at appropriate granularity and support query-efficient pipelines
• Potential to address hallucination in large language models by grounding them in verified knowledge bases

Technology readiness level

The approach is currently at an early-to-mid stage of development. Prior validation in a related drug discovery context demonstrated that LP can successfully learn human-understandable relations and improve predictive accuracy in downstream tasks. The proposed 12-month project will formalise plant-specific data, conduct in silico experiments to learn classification rules segregating physiological effects, and test knowledge transferability across plant species. Deliverables include a logic-based knowledge base and a collection of relational rules for chosen plants. The technology is not yet commercially deployed but is positioned for validation through stakeholder engagement and computational experiments.