OASIS, a self-evolving AI scientist that integrates omics data and literature knowledge for plant stress research.

Abiotic stresses constrain plant growth and crop productivity, yet converting rapidly expanding literature and heterogeneous biological data into testable hypotheses remains difficult. Here we present OASIS, a self-evolving omics-guided AI scientist that links literature-derived mechanistic evidence with omics and other structured data in a traceable workflow for plant stress research. OASIS coordinates six specialized agents for planning, hybrid retrieval, evidence distillation, data interrogation, review, and synthesis. Its knowledge base spans six plant species and four major stress categories and automatically incorporates newly published studies, while the Self-Evolving Experience Learning (SEEL) module distills historical trajectories into reusable procedural rules for experience-based self-evolution. On PlantStressQA, a 200-question expert-curated benchmark, OASIS achieved 84.0/100, exceeding baseline LLMs by 33.9-50.2 points, and showed its largest gains on tasks requiring multi-step integration of literature and data-level evidence. In a rice salt-tolerance case study, OASIS combined Arabidopsis regulatory evidence, orthology, and genetic loci to prioritize five candidates. Loss of OsPP2a function produced salt-response phenotypes accompanied by altered Na+/K+ homeostasis. OASIS is publicly accessible at https://www.oasis.ac.cn.