CropStabDB: data, trained models and validation for a crop protein stability resource across 30 plant species

CropStabDB is a genome-scale resource for plant protein stability and structure-guided protein engineering, covering 1,425,716 proteins across 30 plant species (29 crops and Arabidopsis thaliana, 12 families), each scored by the CropStab Score and decomposed into six interpretable components. This archive contains the complete data, trained models and validation underlying the resource:- master feature table (1,425,716 proteins x 149 sequence-derived features) with CropStab Scores;- the Engineering Atlas of 2,362,229 candidate stabilising substitutions (541,987 high-confidence at ECS >= 0.7) and per-protein summaries;- trained Random Forest and XGBoost models with predictions and SHAP values;- the SQLite database and parquet table behind the cropstab.org web interface;- the 18 validation-layer outputs (AlphaFold pLDDT, N-end rule, ortholog conservation, published families, measured Kd, and more);- manuscript figures and supplementary tables. Source code: https://github.com/Israfil-Hossen/cropstabdbWeb interface: https://cropstab.org