Ah that's really interesting, but confusing - the Rosetta 4.20 application is from May 2020 according to the apps page, so it's from before any Alpahfold/AI was built into Rosetta, other than TRrosetta that was a parallel project, as I understand things. Is there any evidence that these are being pushed to R@H rather than being crunched in-house?

Or maybe they are doing the first stage (the AI bit) in-house and then we are doing the final stage which uses 4.20 on R@H, which might explain the still high volume of work coming through 4.20.

True, the app is over 1 year old. But I have seen the Robetta RoseTTAFold jobs run on my computer with different stages than standard Rosetta.
I haven't seen any sign that the tasks that are actually sent by the Baker Lab/IPD use RoseTTAFold. AFAI can tell, RoseTTAFold is only being run by jobs that are submitted to Robetta from other non-IPD researchers all over the world.

Unless someone tells us what we are actually running, we won't know for sure, but the little info available makes me think that we are indeed running RoseTTAFold jobs through the Rosetta 4.20 app from the submissions to Robetta.

---

Unless someone tells us what we are actually running, we won't know for sure, but the little info available makes me think that we are indeed running RoseTTAFold jobs through the Rosetta 4.20 app from the submissions to Robetta.

The situation is very exciting (at scientific level) but also very confused (see, for example, the latest news of Foldit).
Only admins can dispel the fog

---

The situation is very exciting (at scientific level) but also very confused (see, for example, the latest news of Foldit).
Only admins can dispel the fog

That pretty much summarizes my feelings. I will continue to run them, but at some point I could become convinced that I am just cleaning up old business rather than doing new work.
I would prefer to know which it is.

---

I will continue to run them, but at some point I could become convinced that I am just cleaning up old business rather than doing new work.
I would prefer to know which it is.

From Rosetta@Home Twitter account:

We will be sharing an update about it soon. In short, yes — volunteers are still needed!

---

I will continue to run them, but at some point I could become convinced that I am just cleaning up old business rather than doing new work.
I would prefer to know which it is.

From Rosetta@Home Twitter account:

We will be sharing an update about it soon. In short, yes — volunteers are still needed!

I can't find this reply - can you post a link please?

---

I can't find this reply - can you post a link please?

It's a private message to me.

---

Can't wait for that update.

---

Can't wait for that update.

Rosetta work for GPU's with at least 8GB of VRAM?

---

Can't wait for that update.

Rosetta work for GPU's with at least 8GB of VRAM?

I'm gonna say no lol.

---

Rosetta work for GPU's with at least 8GB of VRAM?

Link

Now, the end-to-end version of RoseTTAFold on RTX 2080 GPUs only takes about 10 minutes to generate skeleton coordinates for proteins with less than 400 residues and can be much more accurate in its predictions as well

---

This is the paper quoted:
https://www.google.com/url?sa=t&source=web&rct=j&url=https://www.ipd.uw.edu/wp-content/uploads/2021/07/Baek_etal_Science2021_RoseTTAFold.pdf&ved=2ahUKEwj48NTc5ovyAhUWahUIHSfUDAQQFjAAegQIAxAC&usg=AOvVaw38GGS9oWPaLckacDXmR4Jp

---

Seems that a LOT of researchers are using RoseTTAFold and/or AlphaFold2.

Structures of core eukaryotic protein complexes
Human mitochondrial protein complexes revealed by large-scale coevolution analysis and deep learning-based structure modeling
Using machine learning to study protein–protein interactions: From the uromodulin polymer to egg zona pellucida filaments
Improved prediction of protein-protein interactions using AlphaFold2 and extended multiple-sequence alignments

The machine learning/AI is deeply changing protein simulations

---

Another interesting paper based on RosetTAFold

We explored the application of the deep learning based structure prediction methods, RoseTTAFold (RF) and AlphaFold (AF), to this problem. Even though RF was originally trained on monomeric protein sequences and structures, it can accurately predict the structures of protein complexes given pMSAs with a sufficient number of sequences. We found that a lighter-weight (10.7 million parameters) RF two-track model provided a good tradeoff between compute time and accuracy: the model requires 11 s (about 100 times faster than AF) to process a pMSA of 1,000 amino acids on a NVIDIA TITAN RTX graphic processing unit, and it can effectively distinguish gold standard PPIs amongst much larger sets of randomly paired proteins

---

Hi everybody, I'm wondering if Rosetta@home still has a purpose if Bakerlabs can now predict proteins much faster with RoseTTAFold. Are there any plans to use this AI on the grid? I think it's much easier to do this in their internal network. But if this is the case, why are there still "normal" Rosetta workunits available? I would guess that the "old" way is ineffective compared to the new approach.

---

Alphafold and RosetTAFold are very good at determining most protein 3d structures, in isolation, which is a massive step forwards. But there is still a huge amount to improve, for example, protein-protein interactions. Or designing new proteins to do a task- I think the cages take are design work, but I might be wrong. Or protein-environment interactions.

So in summary, keep those computers crunching if there is work available.

---

Duplicate...

---

Ok, thank you for the explanation. I'll just rejoined after some years of Rosetta abstinence. Just one more question because I don't have Virtualbox installed. Is there a different purpose in those python workunits compared to the Rosetta 4.20 or Rosetta Mni 3.78 workunits?

---

Yes, the Python tasks run different stuff.
See this September post for some info on the Pythons.

---