User's Email To: "email@example.com"
Date: Wed, 16 May 2012 18:23:16 +0100 Subject: tertiary structure prediction > Dear Dr Zhang, > > We've just found your server and tried analysing our protein sequence > using LOMETS program at > > http://zhanglab.ccmb.med.umich.edu/LOMETS/ > > The results were quite amazing although as new users we are not sure how > the results could be interpreted and which of the many models produced is > the most likely one. > > I understand that prediction is based on similarity in predicted > secondary structures, not on amino acid sequence similarities, correct? > > I'm also wondering if the pdb image generated is that of the reference > protein or of the query protein. > > In case of the latter, how could we identify (locate) particular amino > acid residues (numbers) on the model? > > What is the difference between models predicted by LOMETS and I-TASSER? > > I understand that both programs predict tertiary structures but the > latter is much slower and it takes up to a week to finish. > > Will I-TASSER produce more reliable results? > > Yours sincerely, Replay to the email 1, LOEMTS is a meta-server which have results from multiple threading programs developed from different laboratories. In all the programs, the structure predictions are generated by matching the query sequence to the PDB library in order to identify the best template that have a similar fold to the query. To answer your question specifically, the prediction is not based on amino acid sequence similarity. It is also not only based on secondary structure predictions although secondary structure prediction is an essential consideration. Most of the threading program is based on the secondary structure match plus the alignment of sequnece profiles which was derived from a family of multiple sequence alignments. Some also include structure information from other resources, e.g. solvation and torsion angle etc. 2, There are three type of outputs for 3D structure prediction by LOMETS (see http://zhanglab.ccmb.med.umich.edu/LOMETS/S79). In the column of "Target-template alignments", the data refers to the sequence correspondance between the query and template proteins; In the column of "3-D models from threading alignments", the data represents the C-alpah coordinates which are directly copied from the PDB template structure. There are usually gaps; In the column of "Full-length models by MODELLER", the models are full-length atomic models built by MODELLER by comparative modeling based on the single-template alignment. The models in LOMETS are ranked based on confidence score and the first model is on average better than the second one etc. But it is often found that the lower-ranked models are more accurate than the higher ranked ones. It is wise to consider all the models listed in the first table. If a pdb image is generated, it is on the based of the query protein. 3, The I-TASSER server in general has models with a higher accuracy than LOMETS, since I-TASSER has extensive structure refinement on the templates. This is the reason that I-TASSER takes a longer time to generate the refined models. But the difference is small if there is a close homology identified by LOMETS, which is indicated by the column of "Confidence Score" ("High" usually indicates a close homology). The difference becomes larger if LOMETS cannot identify a close homologous template, where I-TASSER manages to build some part of proteins by ab initio folding.