Hi Randy (and others who replied privately) First off, thanks for taking the time to reply - much appreciated!
This all correlates with what I would have thought. One of the off-board replies mentioned that they remove unstructured/unreliable conformations - I’d guess that these have low confidence or can be pruned out with tools like process_predicted_model and SliceNDice, or even using DSSP to only include elements with identifiable secondary structure. It occurs to me that the problem with misinterpretation of the confidence in the B-factor column could be avoided if there was a field in the mmCIF format to replace “_atom_site.B_iso_or_equiv” with something like “_atom_site.predicted_confidence"; after all, the mmCIF format is ~30 years old so isn’t everyone using it by default (no, don’t bother answering that; learning chemistry in the 1970s I learnt that we should be talking about “ethanal”, “propan-2-ol” etc (“when you work for a chemical company like ICI that’s what you’ll have to call them”), but 50 years later we still buy acetaldehyde and IPA…)? best wishes Harry > On 24 Jan 2026, at 17:35, Randy John Read <[email protected]> wrote: > > Hi Harry, > > Yes, this is easy to answer! There are various papers, some from my group, > about the utility of AlphaFold models in MR. For the specific point of how to > treat the models, you can look at the paper on process_predicted_model > (https://doi.org/10.1107/s2059798322010026 > <https://doi.org/10.1107/s2059798322010026>), which is available in both > Phenix and CCP4. In CCP4, there’s also the SliceNDice tool. > > The side chains are considerably better than random, so they’re left in for > any residues included in the model. But yes, it’s good to prune out low > confidence residues, with 70 being the most common cutoff. > > The trickiest bit is deciding where there might be inter-domain flexibility. > Domain parsing can be done in a variety of ways in process_predicted_model > and SliceNDice, but the human eye is also very good at the required pattern > recognition! > > There’s something I like to reiterate whenever possible, because we keep > seeing people making this mistake. Whatever you do, don’t use the model > straight out of AlphaFold! These have the pLDDT in the B-factor column, on a > scale of 0-100 where 100 is nearly perfect. If you leave them as B-factors, > you will be down weighting the very best parts of the model in the MR > calculation! If you process the model, the pLDDT values will be turned into > B-factors that are low for high pLDDT and vice versa. > > Best wishes, > > Randy > > From: CCP4 bulletin board <[email protected]> on behalf of Harry Powell > <[email protected]> > Date: Saturday, 24 January 2026 at 15:56 > To: [email protected] <[email protected]> > Subject: [ccp4bb] pruning & confidence of AlphaFold models for MR > > Hi folks > > I’m sure that people have thought about this, done this, and maybe even > published on the topic, so this should be an easy one to answer :-) > > I was wondering how people treat AlphaFold models when using them in > molecular replacement? Do they prune them down to poly Ala, do they include > all modelled residues or remove those with low confidence (if so, how low - > pLDDT 70?)? > > Harry > ######################################################################## > > To unsubscribe from the CCP4BB list, click the following link: > https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.jiscmail.ac.uk%2Fcgi-bin%2FWA-JISC.exe%3FSUBED1%3DCCP4BB%26A%3D1&data=05%7C02%7Crjr27%40universityofcambridgecloud.onmicrosoft.com%7C9a680370d247425d6af508de5b613213%7C49a50445bdfa4b79ade3547b4f3986e9%7C1%7C0%7C639048670160264599%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=xIPgNoHXqqATin9xL2qJ6fSY8gDD95VTXvbP0vPdklQ%3D&reserved=0 > <https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=CCP4BB&A=1> > > This message was issued to members of > https://eur03.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.jiscmail.ac.uk%2FCCP4BB&data=05%7C02%7Crjr27%40universityofcambridgecloud.onmicrosoft.com%7C9a680370d247425d6af508de5b613213%7C49a50445bdfa4b79ade3547b4f3986e9%7C1%7C0%7C639048670160291709%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=FyxnNuZ1K1o%2Fy72%2BmxzcAIep3sw4WLzjhEqL56TZuy8%3D&reserved=0 > <http://www.jiscmail.ac.uk/CCP4BB>, a mailing list hosted > byhttps://eur03.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.jiscmail.ac.uk%2F&data=05%7C02%7Crjr27%40universityofcambridgecloud.onmicrosoft.com%7C9a680370d247425d6af508de5b613213%7C49a50445bdfa4b79ade3547b4f3986e9%7C1%7C0%7C639048670160311594%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=Zzj5poOnPPApciMB2vfp2S%2FZndQ3cfPCLIPrvFb4lV8%3D&reserved=0 > <http://www.jiscmail.ac.uk/>, terms & conditions are available > athttps://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.jiscmail.ac.uk%2Fpolicyandsecurity%2F&data=05%7C02%7Crjr27%40universityofcambridgecloud.onmicrosoft.com%7C9a680370d247425d6af508de5b613213%7C49a50445bdfa4b79ade3547b4f3986e9%7C1%7C0%7C639048670160330083%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=g4co4EmBAaJXe0JvtVfD0DSy5XDBYF7KEr1sv26ibgg%3D&reserved=0 > <https://www.jiscmail.ac.uk/policyandsecurity/> ######################################################################## To unsubscribe from the CCP4BB list, click the following link: https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=CCP4BB&A=1 This message was issued to members of www.jiscmail.ac.uk/CCP4BB, a mailing list hosted by www.jiscmail.ac.uk, terms & conditions are available at https://www.jiscmail.ac.uk/policyandsecurity/
