Yang Shen

[ Home | Research | Publications | CAPRI | CV | Links ]

Since January 2005 (Round 6), I have been actively participating in Critical Assessment of PRedicted Interactions (CAPRI), a community-wide experiment that tests protein-docking methods (i.e., methods for modeling protein interactions) in blind predictions.

As a key member of the Vajda Team, I had been continuously developing and applying SDU, a protein docking method that explores and exploits the binding free energy landscape for its global optimization (see the corresponding section in Research), with considerable success. Our team ranked No. 5 among 63 worldwide predictor teams in the 3rd assessment (Rounds 6–12, 2005–2007), and then No. 1 among 64 teams in the 4th assessment (Rounds 13–19, 2007–2009).

I have led my own team since May 2010 (Round 22) and continued on developing protein-docking methods. In the latest assessed Rounds 22–27 right after the 4th assessment (rounds 20 and 21 were only for scorers instead of predictors, so were targets 55 and 56), my team tied at No. 5 among 63 teams. (Last updated on May 10, 2013 according to official summary by Dr. Marc Lensink at the 5th CAPRI Evaluation Meeting.)

Detailed assessments and target information in three rounds are provided in reverse chronological order as follows.

We thank members of the community for many useful comments and feedbacks.

5th Assessment, 2010–2013  (Leader of the Shen Team)

Please be aware that some teams did not participate in all rounds. For instance, the newly-developed SwarmDock server just participated for the last four targets and was the only team that produced at least an acceptable prediction for each of the four. Therefore, please follow a Special Issue of Proteins (December 2013) where teams have more exciting science stories to share than their performance!

1. All assessments are official results according to the CAPRI website. Tied teams are given the same rank and alphabetically ordered.
   
2. For all targets but T47, predictions are classified as * (acceptable), ** (medium), and *** (high). Blank space means that no acceptable predictions were submitted.
   
3. The only, slight exception in classifying predictions was for T47, where the real challenge is the prediction of water-mediated interactions between a given protein sequence and an unbound protein. Here, the classification is * (fair), ** (good), *** (excellent), and **** (outstanding).
4. T48 and T49 both involve binding of hydroxylase and ferredoxin.  The hydroxylase is a hetero-hexamer but there was some confusion among participants that led to some trimer-form predictions.  Therefore, the assessment team decided to evaluate predictions in both forms.  Accordingly, the best model from each team for T47/T48 was reported here, in either hexamer or timer form, as previously treated in targets 40 and 42 in the 4th assessment (see below) where two possible binding modes exist.  
5. T51 was a multi-domain protein GH5-CBM6-CBM13-Fn3-CBM62 among which CBM13  had to be homology-built. Its assessment was performed for three pairwise interfaces.
   
6. T52 was canceled due to available information from the internet.
   

3rd Assessment, 2007  (Key Member of the Vajda Team)

(SDU) indicates that SDU directly contributed to our (best) prediction.