Other features for v2 include template-guidance for #Boltz-2 prediction, auto-contig generation for #RFdiffusion, and models for #HyperMPNN. We have also made bug fixes, improvements to documentation, container updates and more. Details here: github.com/PapenfussLab/proteindj/releases/tag/v2.0.0

Boltz-1x
Chai-1
Protenix
AF2
AF3
Unifold
CombFold
ESMFold
D-I-TASSER
RosettaFold All-Atom
IgDesign
RFantibody
BindCraft
RoseTTAFold2NA
AFCluster
AntiFold
HighFold
BioEmu
DiffAb
AbMPNN
RFDiffusion
HyperMPNN
AntiBERTy
AF-Traj
AbGPT

💬 New preprint evaluation by @arcadiascience.bsky.social of

HyperMPNN – A general strategy to design thermostable proteins learned from hyperthermophiles

https://sciety.org/articles/activity/10.1101/2024.11.26.625397?utm_source=bluesky&utm_medium=social&utm_campaign=preprint_review

#PreprintEvalu

#HyperMPNN - A general strategy to #design thermostable #proteins learned from hyperthermophiles.

Cool stuff! Use this approach to make your favourite protein into #thermostable version!

biorxiv.org/cgi/content/...

Conclusions:
HyperMPNN successfully designs thermostable proteins, demonstrated by a 30°C melting temperature increase in a protein nanoparticle. Limitations include using predicted structures; future directions involve broader applications.

Links:
<a href="https://github.com/meilerlab/HyperMPNN" class="hover:underline text-blue-600 dark:text-sky-400 no-card-link" target="_blank" rel="noopener" data-link="bsky">https://github.com/meilerlab/HyperMPNN

HyperMPNN designed a thermostable I53-50B.4PT1 pentamer variant. Experimental validation showed it remained stable at 95°C, a 30°C improvement over the original 65°C. (pg. 10, Fig. 5)

Retraining ProteinMPNN on hyperthermophile data created HyperMPNN. This model successfully recovered the unique amino acid composition and transferred it to other organisms. (pg. 8, Fig. 3)

Hyperthermophile protein sequences were collected from UniProt, clustered (50% identity), and AlphaFold structures were downloaded and filtered (pLDDT >70). This yielded ~29k high-confidence structures for HyperMPNN training. (pg. 5, Fig. 1)

HyperMPNN - A general strategy to design thermostable proteins learned from hyperthermophiles
HyperMPNN, a retrained ProteinMPNN model using hyperthermophile protein structures, designs highly thermostable proteins. Existing methods struggle with this; HyperMPNN overcomes this limitation ...

HyperMPNN - A general strategy to design thermostable proteins learned from hyperthermophiles

http://biorxiv.org/cgi/content/short/2024.11.26.625397v1

n=1, but IMO the most important result in this paper is the negative result that this training set does no better on designing proteins with high melting temperatures

Interesting that this dataset goes some way to ameliorating ProteinMPNN's borderline-pathological love of lysines and glutamates

HyperMPNN - A general strategy to design thermostable proteins learned from hyperthermophiles https://www.biorxiv.org/content/10.1101/2024.11.26.625397v1

HyperMPNN - A general strategy to design thermostable proteins learned from hyperthermophiles https://www.biorxiv.org/content/10.1101/2024.11.26.625397v1