DigiAra Computationally Designs Plant Mutants for Resistance to Microbial Infection in Arabidopsis

Plant breeding is labor-intensive, requiring repeated cultivation and selection over generations, with few computational tools to support trait design. DigiAra addresses this gap with an AI-based pipeline for designing Arabidopsis thaliana mutants with targeted traits, particularly enhanced microbial resistance. It implements an S3 pipeline—simulation, scoring, and screening—that simulates transcriptional effects of genetic perturbations and infections, scores responses via biological pathway analysis, and screens candidate perturbations at multiple levels. Methodologically, DigiAra introduces a hybrid architecture combining local gene-level interaction modeling with global transcriptional-state modeling to predict perturbation-induced transcriptomic changes. The authors also established a standardized pipeline to curate and harmonize an integrated Arabidopsis–microbe transcriptional dataset of 495 samples from 26 projects. DigiAra achieved a Pearson correlation of 0.49 for predicting gene-expression changes from unobserved perturbations and infections, recapitulated the general non-self response (GNSR), a 24-gene program, and its predicted pattern-triggered immunity pathway scores correlated with bacterial load (Pearson r = 0.57). Genome-wide screening identified 27 gene knockouts predicted to enhance resistance to Pseudomonas syringae pv. tomato DC3000 while limiting growth compromise, 9 of which are supported by published studies. The framework is openly available at https://github.com/youlab2025/DigiAra, and these results position DigiAra as an effective tool for computational plant mutant design.