AF3 ProteinDataset类的get_anchor_ind方法解读

AlphaFold3 protein_dataset 模块 ProteinDataset 类 get_anchor_ind 方法是一个 @staticmethod 静态方法，用来获取“锚定残基（anchor residues）”的索引，目的是在蛋白质序列中被遮蔽（masked）的区域两端找到“已知（known）”的残基，以便后续作为上下文参考。

源代码：

    @staticmethod
    def get_anchor_ind(masked_res, mask):
        """Get the indices of the anchor residues.

        Anchor residues are defined as the first and last known residues before and
        after each continuous masked region.

        Parameters
        ----------
        masked_res : torch.Tensor
            A boolean tensor indicating which residues should be predicted
        mask : torch.Tensor
            A boolean tensor indicating which residues are known

        Returns
        -------
        list
            A list of indices of the anchor residues

        """
        anchor_ind = []
        masked_ind = torch.where(masked_res.bool())[0]
        known_ind = torch.where(mask.bool())[0]
        for _, g in groupby(enumerate(masked_ind), lambda x: x[0] - x[1]):
            group = map(itemgetter(1), g)
            group = list(map(int, group))
            start, end = group[0], group[-1]