抗体-抗原对接¶
抗体-抗原对接是根据抗原结构和抗体的序列/结构来预测抗体-抗原复合物结构的任务。实际应用中,由于算法的预测存在误差,我们会返回前\(K\)个预测结果。
任务难点¶
蛋白质系统的巨大尺寸使得确定两个蛋白质的最佳排列变得非常困难。而且,在对接过程中,两个蛋白质的灵活区域通常会发生构象变化。刚体对接算法无法适应这样的实际情况。
功能亮点¶
- 先进算法: 通过利用最新的几何深度学习技术和大规模结构预训练,GeoAbDock 在抗体-抗原对接方面展现出强大且稳健的性能。
- 支持柔性对接: GeoAbDock 考虑了在对接过程中柔性区域的构象变化。它还支持生成多个候选姿态并对它们进行评分。与刚体对接相比,它更适合实际应用场景。
- 云端大规模预测: 在我们的云平台上支持批量预测和松弛多个序列。同时享受高准确性和高效率!
- 易用:只需要抗原结构和抗体序列即可进行对接。预测的结构会进行聚类,以更清晰地可视化表位分布。
输入¶
要提交抗体-抗原对接任务,请打开项目编辑器并从"结构建模"下拉菜单中选择"抗体-抗原对接"。
- Complex(复合物): 抗原-抗体复合物序列。
- Sequence(序列):抗原-抗体复合物序列。序列中的氨基酸总数不得超过 1200。
>6xc3_C NLCPFGEVFNATRFASVYAWNRKRISNCVADYSVLYNSASFSTFKCYGVSPTKLNDLCFTNVYADSFVIR GDEVRQIAPGQTGKIADYNYKLPDDFTGCVIAWNSNNLDSKVNYNYLYRLFRKSNLKPFERDISTEIYQA GSTPCNGVEGFNCYFPLQSYGFQPTNGVGYQPYRVVVLSFELLHAPATVCGPK >6xc3_H QMQLVQSGTEVKKPGESLKISCKGSGYGFITYWIGWVRQMPGKGLEWMGIIYPGDSETRYSPSFQGQVTI SADKSINTAYLQWSSLKASDTAIYYCAGGSGISTPMDVWGQGTTVTVSSASTKGPSVFPLAPSSKSTSGG TAALGCLVKDYFPEPVTVSWNSGALTSGVHTFPAVLQSSGLYSLSSVVTVPSSSLGTQTYICNVNHKPSN TKVDKKVEPKSC >6xc3_L DIQLTQSPDSLAVSLGERATINCKSSQSVLYSSINKNYLAWYQQKPGQPPKLLIYWASTRESGVPDRFSG SGSGTDFTLTISSLQAEDVAVYYCQQYYSTPYTFGQGTKVEIKRTVAAPSVFIFPPSDEQLKSGTASVVC LLNNFYPREAKVQWKVDNALQSGNSQESVTEQDSKDSTYSLSSTLTLSKADYEKHKVYACEVTHQGLSSP VTKSFNRGECS
- (上传序列): 点击 "" 等结尾。参见以上示例。}" 按钮上传您自己的抗原-抗体复合物(一个 .FASTA 文件)。文件中只能存在一个抗原-抗体复合物,各链的 FASTA 标签必须具有相同的前缀且以 "_{chain_id
- Sequence(序列):抗原-抗体复合物序列。序列中的氨基酸总数不得超过 1200。
- Job Name(任务名称): 任务的名称。请注意,任务名称必须在项目内唯一。
模型 & 参数¶
您可以使用我们自研的 GeoAbDockv4 模型来运行本任务。此模型的参数如下:
- Mode(模式): 您可以在 “fast(快速)“ 和 ”accurate(精确)“ 之间选择,前者速度更快,后者会进行更多对接姿态的采样,对接成功率更高。
结果¶
在文件与任务管理器面板中点击任务结果,查看任务结果。 结果存储在 CSV 文件中,可以通过点击结果表格右上角的""按钮下载。
结果表格包含以下列:
- rank(排名): 从 0 开始的样本排名,按照 ilDDT 评分降序排序。
- score(置信度打分):预测结构的 wpTM 打分(范围从 0 到 1),表示对预测的置信度。它是通过 \(\text{wpTM} = 0.8 \times \text{ipTM} + 0.2 \times \text{pTM}\) 计算得到的。
在最右边的列中,您可以点击 "" 按钮查看复合物结构。