<p>Hyperparameters used for training various models on five different protein datasets.</p>

The table lists the batch size (32, 64, 128, 1024, 2048), learning rate (0.005, 1 × 10 – ³, 1 × 10 ⁻ ⁴), number of dense layers (1, 2, 3), and number of nodes for NN models (100, 200). For CNN models, the number of convolutional layers and filter sizes are reported, while for GCN models, the graph thresholds (in Å) are specified. If mutational or positional splits used different hyperparameters, they are indicated separately within the corresponding cell.
(XLSX)

本表格列出了神经网络（Neural Network, NN）模型的批量大小（取值为32、64、128、1024、2048）、学习率（取值为0.005、1×10⁻³、1×10⁻⁴）、全连接层（dense layers）数量（取值为1、2、3）以及节点数（取值为100、200）。针对卷积神经网络（Convolutional Neural Network, CNN）模型，表格会报告其卷积层数量与滤波器尺寸；而对于图卷积神经网络（Graph Convolutional Network, GCN）模型，则会明确其图阈值（单位：埃（Å））。若采用突变拆分或位置拆分时使用了不同的超参数，则相关超参数会在对应单元格内单独标注。（XLSX）