Predictions from OrgNet+: towards robust protein stability prediction with convolutional neural networks

1) Original Benchmarks (Non-augmented):Predictions for OrgNet+, OrgNet, and other structure-based methods on the S461 and S669 datasets, evaluated in direct and reverse mutation settings.OrgNet predictions are from Buyanov et al. (2025). Predictions from other methods are sourced from Iqbal et al. (2022, PROST).

2) Conformationally Augmented Benchmarks:Predictions for five conformational ensembles (Backrub, Boltz-2, iMod, MD, NOLB), reported per-ensemble for both direct and reverse settings.Evaluated models include: OrgNet+ (trained on the full augmented dataset), OrgNet+ (trained on individual S2648-derived ensembles), OrgNet, and RaSP.RaSP model and prediction scripts were obtained from: https://github.com/KULL-Centre/_2022_ML-ddG-Blaabjerg/

1) 原始基准集（非增强版）：针对S461与S669数据集，在正向突变与反向突变设置下评估得到的OrgNet+、OrgNet及其他基于结构的方法的预测结果。其中OrgNet的预测结果源自Buyanov等人（2025）的研究，其余方法的预测结果来自Iqbal等人（2022，PROST）的工作。
2) 构象增强基准集：针对五个构象集合（Backrub、Boltz-2、iMod、MD、NOLB）的预测结果，按每个集合分别报告正向与反向突变设置下的结果。本次评估的模型包括：在完整增强数据集上训练的OrgNet+、基于单个S2648衍生构象集合训练的OrgNet+、OrgNet以及RaSP。RaSP模型与预测脚本获取自：https://github.com/KULL-Centre/_2022_ML-ddG-Blaabjerg/