由欧洲生物信息学中心组织的国际蛋白质-蛋白质相互作用预测竞赛，于2010年1月4日至31日举行第20轮。这轮竞赛的两个题目T43和T44均来自美国Washington University的David Baker实验室。这次的题目与以往19轮的题目形式不一样，以往都是在给定已知会相互作用的2个单体的基础上预测复合物结构，而这次是给出了众多人工计算设计的复合物，不确定是否真正相互作用，也不确定复合物三维结构是否正确。
这次的2个题目简介如下：
T43有21个A、B两链蛋白质复合物三维结构。其中有20个是人工计算设计的，与天然晶体结构偏差较大。另一个是晶体解析的天然结构。21个复合物的A链实际上是6个不同的蛋白质，B是18个不同蛋白质，所以总共涉及24个蛋白质。我们的任务就是把这其中唯一的天然结构挑出来。
T44也有21个A、B两链蛋白质复合物三维结构，这里的所有复合物都是人工计算设计的，有待进一步做生化和晶体衍射实验确认是否有复合物中的A、B两个蛋白质真正相互作用，甚至结构预测是否正确。21个复合物的A链实际上是3个不同的蛋白质，B是21个不同蛋白质，所以总共也涉及24个蛋白质，只不过与T43的蛋白质不一样。我们的任务是挑出其中有可能真正相互作用的复合物结构，具体数目不确定，可能一个都没有，也可能有多个。
我们昨天提交了预测结果，但最终正确与否还需要等待Baker小组和CAPRI委员会的评判。待评判结果下来后，我将具体分析我们自己预测方法的长、短处。
T43的结果：
All the 21 models of Target 43 have been evaluated as the following steps:
1. Divide the 21 models into 6 groups according to chain A in the models. Models in one group have the similar chain A.
G1: 1-4; G2: 5,6,7; G3: 8,9,10; G4: 11,12,13; G5: 14-18; G6: 19,20,21.
2. Predict binding site patch residues for all the monomers A and B. At the same time, search the homologue protein sequences for all the chains A and extract experimental literatures to analyze the related pdb 3D structures.
At this step, we want to make clear the aim of every designed model and to compare the designed models with the experimental 3D structures.
We think one of the aims is to design an protein inhibitor for the chain A of G1 (models 1-4), which is the protein interleukin 23.
3. Remove the models with an complex interface area less than 1200 Å2.
7, 9, 12, 15, 16, 17, 19, 21
4. Remove the models with a gaussian correlation facotr between the chains A and B less than -0.22.
6, 8, 10, 11, 13, 18, 20
5. Remove the models with wrong binding site patch residues according to prediction and experimental information.
5
6. Remove the models with wrong axes orientation between the chains A and B.
1, 4, 14
7. Remove the models with wrong correspondence between the binding site patches compared with the experimental pdb structure 3DUH.
2
At last, we obtain the model 3. So we submit model 3 as the result.
T44的结果：
All the 21 models of Target 44 have been evaluated as the following steps:
1. Divide the 21 models into 3 groups according to chain A in the models. Models in one group have the similar chain A.
G1: 1, 2, 3; G2: 4-7, 15-21; G3: 8-14.
2. Predict binding site patch residues for all the monomer chains A and B. At the same time, search the homologue protein sequences for all the chains A and extract experimental literatures to analyze the related pdb 3D structures.
At this step, we want to make clear the aim of every designed model and to compare the designed models with the experimental 3D structures.
We think one of the aims is to design an protein inhibitor for the chain A of G3 (models 8-14), which is the protein hemagglutinin.
3. Remove the models with an complex interface area less than 1200 Å2.
9, 10, 12, 14, 15, 16, 17, 19, 20, 21
4. Remove the models with a gaussian correlation facotr between the chains A and B less than -0.22.
7, 11, 18
5. Remove the models with wrong binding site patch residues according to prediction and experimental information.
1, 2, 3, 4, 5, 6
6. Remove the models with wrong axes orientation between the chains A and B according to the pdb structure 3GBM.
8
At last, we obtain the model 13. So we submit model 13 as the result.
