#GoodThingsComeToThoseWait Sorry it's taking us a while. Promise that the wait will be well worth it! #OpenFold3 Love all the great work that's already been done #HelixFold3 #Ligo #Chai1 #Protenix #Boltz1

I added support for Boltz1 so we can calculate ipSAE and other scores for pairwise protein-protein interactions on AF2, AF3, and Boltz1 models. Will add nucleic acids eventually. github.com/DunbrackLab/...

If you use Boltz1/2, BioEmu, Chai1, or other MSA-dependent models, you’re likely using our ColabFold server. Please be considerate! Avoid large submissions across many IPs instead generate the MSA locally. Our server is an old-timer from 2014 and can’t handle that load.

Finally I could build all closed environment for structure prediction with colabfold, mmseqs2 and boltz1.
And the pipeline worked well. 🍻

We are folding a genome this weekend with Boltz1 which has been previously folded with ESMfold and also some with AF3 and AF2. I’ll let you know how it goes.

New Title Alert: af_analysis- is a Python package for analysis of AlphaFold protein structure predictions from AlphaFold 2/3, ColabFold, AlphaFold-Multimer, AlphaPulldown, Boltz1, Chai-1, and MassiveFold.

Learn more here: buff.ly/BeBrAei

#SBGrid #Software #Analysis

Homology retrieval grounds ML systems to produce reliable predictions. MMseqs2 is already used in Boltz1/2, BioEmu, MSA-Pairformer, Chai-1, BioNeMo, Proteinx, etc. MMseqs2-GPU can enable these and next-gen models to integrate fast homology retrieval for end-to-end GPU inference. 3/n

I missed that… thanks for highlighting this difference!

I also wonder how different it would be the comparison with ipSAE min/max from Boltz2 with Boltz1/AF3.

I wish I had some GPUs to test it myself….

For antibodies, AF3 is actually significantly better with equal sampling. In our dataset. We are a bit slow writing up the paper about that, but here is the key figure. For some strange reason, boltz1, is really bad, likely we had a bug. Y axis is DockQ, X number of samples.

I got the code to process Boltz1 files. Not sure if processes Boltz2 outputs properly. But I have not tested to see if ipSAE scores Boltz1 or Boltz2 output in a useful way. I'd be happy for someone to try it and share their results.

Вчені представили AI-модель Boltz-1 для прогнозування біомолекулярних структур https://root-nation.... #AI #Boltz1 #Біомолекули #ШтучнийІнтелект #MIT

Ithe improvements over Boltz1 s due to one target L3000 Autotaxin

Cluspro was better on l1000

#casp16

Very nice - I hoped this might happen. I notice MPNN struggles with particular targets. In these cases, as an alternative to avoid adversarial sequences, I've been screening relaxed trajectories with Boltz1 or chai1 to recover AF designed sequences that have <2 Å RMSD, low PAE etc. Any thoughts?

Boltz-1 Democratizing Biomolecular Interaction Modeling | bioRxiv #boltz1 #modeling #protein #complex #interaction #prediction #drugdiscovery #deeplearning

Has anybody noticed how #Boltzgen does not output pLDDT in the ADP column of the output .cif files in the way Boltz1/2, AF2/3, Chai1 etc do? Especially weird since it uses Boltz2 to generate these models, has it been taken out to speed up calculations? Preprint doesn't mention it.

Lewis Martin's script to run Boltz on Modal: gist.github.com/ljmartin/eae...