MAV: co-variance analysis

[olibclarke@…]

Completely different suggestion which may not be feasible and would involve adding functionality - I really like the access to sequence analysis in chimera. One thing which I can't find, and which would be great to have for both structure analysis and validation purposes, is some kind of interface to calculate and display sequence covariance information (a la Gremlin, EVfold, freecontact etc) - this kind of information is great for identifying sequence register errors when building structures at low resolution, as well as for identifying functionally important residues in a structure, but there are not really many easy ways to visualise this information, so incorporation of such a tool into Chimera would fill a gap (particularly if it incorporates multiple modes of analysis - e.g some residues have a extremely conserved binary contact with another residue, whereas others have conserved contacts with several residues - this is where I could see a network graph display being very useful).

--and--

Yes - the only ones I have found to be useful are GREMLIN and EVFOLD, but they only work well if you use a massive sequence alignment (usually >5 sequences per aa), which both servers will happily generate if enough sequences are available but which is cumbersome to generate manually. In both cases, however, visualization of the results is lacking. I believe there is an open source implementation of the Gremlin method called CCMpred (​http://bioinformatics.oxfordjournals.org/content/30/21/3128.long), though you are right, maybe it is more effort than it is worth.

[Elaine Meng]

Does Jalview interface to any covariance methods?  Might be worth talking to our friends in Dundee about this, as they certainly have more experience handling massive sequence alignments than we do.  

Secondly, sequence covariance is also a step within the AlphaFold2 calculations and I wonder whether such intermediate data could be extracted and simply shown in another program like Jalview or ChimeraX.  

My sense is that we probably would not want to try to come up with our own method, given that bioinformatics gurus have been working on this area for decades.