AF3 Cropper类解读

AlphaFold3 protein_datamodule 模块 Cropper 类用于裁剪蛋白质输入数据的 PyTorch nn.Module 模块。在 ProteinDataModule 的数据预处理流程中，它是 transform 组合的一部分，主要作用是将蛋白质的相关特征裁剪为固定长度（如 384 residues），以适配下游模型的输入要求。

源代码：

class Cropper(nn.Module):
    """A transformation that crops the protein elements."""

    def __init__(self, crop_size: int = 384):
        super().__init__()
        self.crop_size = crop_size

    def forward(self, protein_dict: dict):
        """Crop the protein
        :param protein_dict: the protein dictionary with the elements
         - 'X': 3D coordinates of N, C, Ca, O, `(total_L, 4, 3)`,
         - 'S': sequence indices (shape `(total_L)`),
         - 'mask': residue mask (0 where coordinates are missing, 1 otherwise; with interpolation 0s are replaced
                   with 1s), (total_L),
         - 'mask_original': residue mask (0 where coordinates are missing, 1 otherwise; not changed with
                              interpolation), `(total_L)`,
         - 'residue_idx': residue indices (from 0 to length of sequence, +100 where chains change),
                            `(total_L)`,
         - 'chain_encoding_all&#