PlantEssentialGenePredictor: models, processed features, genome-scale predictions and independent-validation evidence

This dataset contains the trained model artifacts, processed feature matrices, fixed label tables and genome-scale prediction results associated with PlantEssentialGenePredictor, a plant essential-gene prioritization framework for Arabidopsis thaliana and rice (Oryza sativa). The archive includes single-species Arabidopsis, single-species rice and joint Arabidopsis-rice models, deployable feature-profile models for different annotation combinations, processed 6751-dimensional feature matrices, fixed train/validation/test labels, and released prediction tables for unknown Arabidopsis genes and rice genes. Source code, documentation and the Streamlit web application are available at:https://github.com/tianluyao1024/PlantEssentialGenePredictor Version v1.2.3 adds an independent-validation evidence supplement containing Europe PMC source-screening ledgers, curator-checked direct loss-of-function records, zero-overlap audits, external-cohort eligibility reports, candidate evidence cards and Figure 7 source data. The external phenotype cohort does not meet the preregistered quantitative threshold of at least 30 genes per species and at least 10 genes per class. Therefore, no external AUC, AUPRC or threshold-based performance estimate is reported in this release. Raw phenotype databases and pretrained ESM2, ProtBERT and ProtT5 weights are not redistributed.