Predictive breeding platform for drought-tolerant cacao genotypes using eco-physiology and machine learning

Overview

Climate change is driving more frequent and severe drought events, posing a major challenge to cacao production. Theobroma cacao is inherently inefficient at controlling water loss, and drought tolerance varies significantly across genotypes depending on morphological and physiological traits. This solution offers a predictive breeding platform that merges eco-physiological modeling with machine learning to forecast drought tolerance in cacao genotypes, enabling faster and more accurate selection of tolerant varieties for grafting and cultivation in drought-prone neo-tropical regions.

The platform addresses a critical bottleneck in perennial crop breeding: the long evaluation cycles traditionally required to identify drought-tolerant genotypes. By using existing phenotypic data as a calibration dataset, the approach enables breeders to predict the tolerance phenotype of untested genotypes, including at the rootstock-scion interaction level, without years of field trials.

Technical specifications

Core approach:

• Integration of eco-physiological indices with machine learning models as reliable proxies for selection indices within breeding programs
• Calibration using a training population of 14 cacao families (full sib and open-pollinated) evaluated over two years under greenhouse conditions across two water regimes
• Assessment of eight morphological and physiological traits to identify contrasting genotypes
• Extension of models to account for the complexity of rootstock-scion interaction under water deficit conditions
• Validation across cacao seedlings at local nurseries

Key features:

• Replaces traditional ABLUPs with modern, data-driven predictive tools
• Enables forecasting of drought tolerance for untested genotypes based on genealogical information
• Supports polygenic adaptation breeding in perennial tree crops
• Designed for targeting neo-tropical drought-prone localities

Technology readiness level

The technology is at an advanced validation stage. The research team has already completed two years of greenhouse screening across 14 families and eight traits under two water regimes, identifying three promissory full sib families and two highly susceptible families. This contrasting dataset provides a robust training population for model calibration. The next phase involves developing explicit eco-physiological indices, calibrating machine learning models, extending predictions to rootstock-scion interactions, and validating predictions across cacao seedlings at local nurseries through AGROSAVIA's in-kind contributions. The platform is positioned to accelerate breeding cycles and reduce time-to-market for drought-tolerant cacao varieties.