<b>CryoNet.Refine benchmarks</b>

We evaluated CryoNet.Refine across 900 cryo-EM structures spanning diverse molecular types and imaging modalities (resolutions 1.8 to 8.0 Å). The evaluation sets include 225 AlphaFold 3–predicted starting models, 325 pre-2018 PDB deposits, 350 post-2018 PDB deposits and 50 half-map PDB deposits (refined against first half-map).

我们针对900个涵盖不同分子类型与成像模式、分辨率范围为1.8至8.0埃的冷冻电镜（cryo-EM）结构，对CryoNet.Refine开展了评估。本次评估数据集包含225个由阿尔法折叠3（AlphaFold 3）预测得到的初始模型、325条2018年前存入蛋白质数据库（Protein Data Bank，PDB）的结构条目、350条2018年后存入PDB的结构条目，以及50条基于第一半密度图完成精修的半密度图PDB存档结构。