Much ado about Foldit

Recent ado about FoldIT success for prediction of the structure of monomeric retroviral protease (PDB ID: 3HVP) stack me into the question: We didn't solve the structures of proteases of the most common viruses yet? I can't believe my eyes...so I begin small investigation.
1. Let's take a look at the seqeunce similarity of the solved monomeric retroviral protease - 3HVP
2. Does it have homologs? Of course! Seq.Similarity list 395 structures in the PDB with sequence similarity 95%.
3. I have randomly took the 2HS1 structure for protein superposition. RMSD of the Calpha-atoms alignment is 1.6 Angstrom. 2HS1 Structure is published in 2006 (the most old between them is 1989).
4. So the question arised - why authors can't model the protein by homology modeling, molecular dynamics optimization, etc. There is a lot of thechniques exist to model structure based on bad NMR data. So, the gamers from Foldit solved the problem better than the authors do later the same year. Seems very strange for me...


An example of ROC curves plotting with ROCR

Decided to start github with ROC curve plotting example. There is not a one ROC curve but several - according to the number of comparisons (classifications), also legend with maximal and minimal ROC AUC are added to the plot. ROC curves and ROC AUC were calculated with ROCR package.

What else should be added to the plot for ease of understanding?