 Once the models are built, we are on the Models tab. Click on the picture to select the first model. We see once more the oligomeric state and the ligands. The arrow expands the view so that we can see the exact interactions. If we move the mouse over the entry, the amino acids involved in the interaction are highlighted in the structure. The global model quality estimate is an updated version of the one shown in the template selection, because it also takes the Q-Mean score into account. Now, what is the Q-Mean score? Well, first of all, we built a model. This means we do not know what the correct structure really looks like. But we still want to know how good our model is. Swiss model uses the Q-Mean score for quality estimation. It is easiest to explain if we click here to extend our view and have a look at this comparison plot. OK, so how can we estimate the quality of a model or anything if we don't know the answer? Well, we could describe some properties of known structures and then see how well these properties are represented in our model. And that is exactly what is shown here. On the x-axis, we have the length of the protein structure and in the y-axis, the score of these properties. Every dot in the plot represents one structure from the PDV database and the star is our model. The darkest color represents the structures with a z-score below plus minus one. If you are not familiar with z-scores, but familiar with statistics, you can think of within one standard deviation of the mean in a normal distribution. In this case, our model behaves very similar to the experimentally determined PDV structures. Now, what exactly does the Q-mean score consist of? Well, you can see the components to calculate Q-mean score in these bar plots below the Q-mean z-score here. The first compares the interaction potential either only between the C-beta atoms or, in the second case, all atoms. The solvation potential calculates the solubility of the protein. For the torsion angles, you can think about the Ramachandran plot, except it does not only look at the amino acid itself, but also takes into account the two neighboring residues. Again, represented as z-scores in comparison to experimental structures. This single value of Q-mean is a good indication of how reliable the model might be. However, answering the question, is it a good or bad model, is more complicated than this. Values below minus four are an indication that the model is very unreliable. However, usually a more important question is which parts of the model are reliable and which ones are not. And this can be seen in the local quality estimate. The plot gives a quick overview. As a rule of thumb stretches below 0.6, hint towards regions with low quality. If we expand the model template alignment and click on the gear icon, we see that we have a new option how to color it. If it is not yet set to Q-mean 4, choose it. This colors the sequence and the structure according to the local quality estimate. We see right away that the beta barrels and the protein-protein interface is modeled with high confidence. The same is true for the binding site of the copper and to a lesser extent to the zinc binding site. While these loops seem to have a lower confidence. This is also reflected when we select several models. See how the core aligns very well but the loops are different? This is quite expected. If we go back to the template alignment view, we see that both the structure and the sequences differ in the loop regions. The difference in the sequence alignment means that the modeling engine cannot rely as much on the template for modeling here, which indirectly explains the drop in quality. If we select the model from the other cluster, we can also immediately see what this other cluster was. This chain aligns perfectly with the other models, but the interface between the two chains is a completely different one. We already know that we are not interested in this model, otherwise we should go to the template and investigate it further. Also, once you have finished, don't forget to download your models, or better yet, grab the whole project. This not only saves all models but also the template search and the quality estimates.