Thread by @cheshireminima

Since the publication of #AlphaFold2 and #RoseTTAFold and now that the tools and models have been made accessible, there has been an avalanche of attempts to solve old crystal structures. This thread covers tips from the #PhaserTeam for doing #MR with these models. (1/...)

The first thing to be aware of is that the B-factor fields contain measures of confidence in the correctness of the prediction, not actual B-factors. This means: 1) we can use that confidence to trim the model and 2) we need to convert that to an appropriate B-factor. (2/...)

Phaser will take those B-factors and use them to weight the different parts of the model. This can improve your chances of success with the model. (3/...)

If your model comes from the #RoseTTAFold server, it will have an estimated rmsd error in A in the B-factor column. They already give the option to obtain versions of the model from which atoms with an error estimate over some value have been trimmed. (4/...)

For #MR in #phaser you should process the #RoseTTAFold model. From the latest nightly build from #phenix, you can use our tool phenix.voyager.rmsd_to_bfactor to process those error estimates to B-factors and also optionally trim the model. (5/...)

If your model comes instead from the #AlphaFold database at the #EBI, or from their Colab notebook, the metric on their B-factor column is the LDDT. This is the score that is shown as quality metric, and is a percentage with 100% being best. (6/...)

@RandyJRead has done some exploration of how LDDT empirically relates to RMSD, and with that and our previous formula to convert from RMSD to estimated B-factors, now we also have a small tool, phenix.voyager.lddt_to_bfactors available in the most recent nightly build. (7/…)

You really want to process the #AlphaFold models, because if the LDDT scores are treated as B-factors, the best parts of the model will get the lowest weight in the #MR calculation! (8/..)

@sokrypton was kindly told me that if you run #RosettaFold locally or in a notebook the LDDT will be in the range from 0 to 1 instead of 0 to 100, so we will update our script to deal with that too. I'll keep you posted! (9/...)

Also @BenButt2 has told me locally run AlphaFold models will have an empty bfactor column and the IDDT will be on a different file!

The tool was updated to check if lddt is in percentage or fractional form. You can find on the very latest phenix nightly builds available on phenix-online.org/download/night…. Even if tagged as incomplete on some platforms they will work on the others. PM me if issues!