Phenix capabilities where ChimeraX interaction would add great value

• From: Tristan Croll
• April 21, 2020

Hi Tom,

I do admit that a purely COVID-19 grant will be a bit of a tough sell, but it's worth a shot. In the larger scheme of things, yes: the goal would be all about improving how quickly and easily we can build high-quality models - particularly into low/medium resolution density.

The things I can think of that could be directly applicable to COVID-19:

• ligand parameterisation (already discussed)

• initial ligand docking into density (that is, to get the initial coordinates for ISOLDE to work from) - Phenix has a tool called LigandFit that will search your density (crystallographic or cryo-EM) for candidate blobs and then flexibly fit a ligand, but it's purely menu-driven with no model visualisation front end nor mechanism to specify where in the density to look. That makes it quite slow and a bit unintuitive to use. On the other hand, ChimeraX has all the model visualisation tools and API to specify a location and output a local map/model fragment, but no flexible-docking tool. Put them together, and it could make for a very nice "place ligand here" tool. I've asked Tom Terwilliger about this one, and he thinks the Phenix side should be very straightforward to implement.

• building in missing segments. I'm just about at the point of having an "extend chain by one residue" tool that I'm happy with exposing to the public interface, but Phenix has the necessary tools to extend by longer fragments at a time (but like all other tools of its nature, makes increasingly serious mistakes at lower resolution). Turning it into an interactive tool would have significant benefits in both speed and quality. This is unlikely to be important in the context of any of the existing COVID-19 structures, but there are still quite a few viral proteins whose structures haven't been experimentally determined. Making these easier to build well would be valuable.

• validation and deposition. Phenix has a much more comprehensive suite of validation tools than ISOLDE - particularly useful things that I don't have yet include bond length, angle and chirality validation, nucleic acid geometry, local fit-to-density outliers, and listing of unmodelled blobs.

• validate ligand linkage. One that probably shouldn't be mentioned on grant applications, but one where I think a good graphical front end could be immensely valuable: the current policy of both phenix.refine and phenix.real_space_refine is to ignore all LINK and CONECT records on the basis that historically these have been very unreliable (e.g. due to various modelling packages either leaving them out entirely or not updating them after edits). Instead, the default is to automatically create links based on the input geometry - which would be great if it were foolproof. Sadly, it's not - on multiple occasions now I've found deposited structures where it's turned bad clashes into nonsensical protein-ligand bonds which have then sneaked past all validation and peer review. Just the other day, a colleague asked me to look over his structure (in its second round of review for Nature, and held up only because a reviewer wanted some extra analysis of a key site)... half a dozen ligands with wrong bonds to the protein, pulling things out of density. An obvious solution to this issue would be a simple interactive check: "we think these sites look like they might be bonds. Please check and confirm"... with a clickable table that takes you to each spot and an "is bond" checkbox to decide.

Hope this helps.

Tristan