Author : Deepa Bhadana and G. Arulkumaran

Plant diseases, especially those that can be transmitted through seeds like Fusarium wilt, remain a daunting threat to agricultural productivity worldwide. Very often, traditional methods of diagnosis depend on visible symptoms and manual laboratory tests, causing delays in accurately identifying plant diseases. The era of genomic technologies has emerged, and CRISPR-based markers have begun to yield useful information regarding genetic resistance traits. Nevertheless, correlating such high-dimensional genomic data with phenotypic traits for accurate prediction continues to pose a challenge. This study proposes a hybrid deep learning–machine learning framework, based on an Autoencoder (AE) and XGBoost, to predict resistance against Fusarium wilt using genomic and phenotypic data related to CRISPR enhancement. The Autoencoder extracts features unsupervised and reduces dimensionality to capture complex, nonlinear patterns while filtering noise effectively. The compressed latent features are classified by the XGBoost algorithm, which implements gradient boosting techniques and is robust toward structured and imbalanced data. Genomic and phenotypic input data are horizontally concatenated (early fusion) into one single training matrix to ensure both types of data are used jointly during learning. Hyperparameter optimization using Bayesian Optimization maximizes classification accuracy and minimizes loss. The model, therefore, has a far better resistance prediction accuracy than existing models that give weight to biological interpretability and computational efficiency. With the boost of CRISPR-based markers, the developed model becomes a significant instrument for early detection, breeding decisions, and sustainable plant disease management.

Keywords: Plant disease prediction, Fusarium wilt, CRISPR-based genomics, Autoencoder, XGBoost, Disease resistance classification