The project done here shows a great example of how an antibody with an unknown structural characterization can be computationally determined to some degree. In the case of NRP2 and the first antibody 7LYV the RMSD values of both predicted structures were accurate in their determination. The only problem was with antibody 7M1C as this antibody did not fully match the reference structure however it did align well to where the antibody should be located in an experimental wet lab structure. As mentioned previously the CDR loops of the crystal structure of antibody 7M1C where not fully determined and so an accurate assessment could not be done without rebuilding the antibody from scratch with the sequence and then using Rosetta to snugdock the antibody as snugdock is specifically made for antibody loop modeling. The reason it was not used in this experiment was because there were some technical difficulties between the published .fasta file and the Rosetta antibody modeling package.
Overall the results show a good predictive homology model if the crystal structure did not exist. Although not fully noticeable right away, it is clear that 7LYV blocks infection of Human Cytomegalovirus by binding to the site where the calcium atom of NRP2 would coordinate, which is consistent with what the paper this project was based on describes.
It is important to note that the mechanism of action for antibody 7M1C is similar to 7LYV. The paper from which this project was inspired from creates a three dimensional reconstruction of the cryo em maps to find two binding sites for NRP2, 7M1C blocks the binding of NRP2 by binding to its binding site.
One more note. In the start Home portion of this website it was mentioned that this could be done within an hour. This experiment was done with Docking 2 initially on Rosie which took longer due to the production of 1,000 structures and was less accurate. Initially this took weeks to be done due to the inability to use snugdock as building the antibody with cdr loop definitions was beyond the scope of the abstract. Later on Docking 3 was used for this experiment as a back up, however this ended up processing faster due to only 100 predicted models being computated and as well it was more accurate. Docking 3 on the rosie server to complete this total project took 3 hours compared to the initial weeks. Showing the power of current computational software to use in your experiments.
Now you may ask, what do I do with this data? The answer is simple, by predicting which residues interact with your antigen, you could make your antibody more specific, or have higher affinity, or be more thermally stable, ect.. The prediction of the Antibody-Antigen complex can lead to faster antibody design and development.