Dayhoff is an Atlas of both protein sequence data and generative language models — a centralized resource that brings together 3.34 billion protein sequences across 1.7 billion clusters of metagenomic and natural protein sequences (GigaRef), 46 million structure-derived synthetic sequences (BackboneRef), and 16 million multiple sequence alignments (OpenProteinSet). These models can natively predict zero-shot mutation effects on fitness, scaffold structural motifs by conditioning on evolutionary or structural context, and perform guided generation of novel proteins within specified families. Learning from metagenomic and structure-based synthetic data from the Dayhoff Atlas increased the cellular expression rates of generated proteins, highlighting the real-world value of expanding the scale, diversity, and novelty of protein sequence data

Researchers at Tsinghua University created a new system called DrugCLIP, that can screen drug molecules against human proteins at a speed that makes traditional methods look ancient.

> DrugCLIP uses deep contrastive learning to turn both molecules and protein binding pockets into vectors and match them almost instantly.

> It screened 500 million molecules across 10,000 human proteins, covering half of the entire human druggable proteome.

> The system completed 10 trillion molecule protein evaluations in a single day, roughly 10 million times faster than classic docking simulations.

> They used AlphaFold2 to generate protein structures and then refined binding pockets with a custom tool called GenPack.

> The model even identified compounds for TRIP12, a protein linked to cancer and autism that has resisted traditional drug-targeting approaches.

Our mission is clear: to advance the frontier of AI capabilities in biology through open science and make them universally accessible to every scientist working towards a healthier, more sustainable future.

Today, we are excited to announce some major steps in this direction: the unveiling of our new small-molecule and protein design agents, the launch of Boltz Lab, our seed round, and the announcement of a partnership with Pfizer.

A next-generation folding model that scales up model capacity through width scaling and large-scale data distillation. We also provide SeedFold-Linear, a more efficient variant with linear triangular attention. Both models achieve state-of-the-art results on FoldBench, outperforming AlphaFold3 on most protein-related tasks.

Multi-algorithm protein structure prediction using Modal serverless infrastructure: semi-vibe coded frontend tool for the protein folding tools Chai1, Boltz2, AF2, Protenix(-mini).

Not in how long each batch of tasks lasts, not in Boinc scheduling, not in runtime, not in credit received and not in the amount of processing lost when the PC is shutdown or rebooted or crashes.

There are no advantages to having a shorter runtime.
Not in how long each batch of tasks lasts, not in Boinc scheduling, not in runtime, not in credit received and not in the amount of processing lost when the PC is shutdown or rebooted or crashes.
I'm not saying there are advantages in all those factors - they're neutral at worst - but whatever differences there may be are advantages, either to you personally or to all other users of Rosetta.

Everyone should make this change to their default runtimes imo - no ifs or buts

This is a guide is intended for scientists interested in getting into computational protein design, and in silico scoring.

ProFam-1 is a 251M-parameter autoregressive protein family language model (pfLM), trained with next-token prediction on concatenated, unaligned protein sequences drawn from the same family.

Is that right? Maybe I'm getting mixed up with when each decoy is completed.

Edit: Now finished and 8 decoys completed - very likely checkpointing after each decoy. Not the worst, but not the best either.

Not only that, but they checkpoint very infrequently too - maybe once every 3hrs at best

If you recall, in the large batch of work issued about 6 months ago, there was a problem in the job turning queued tasks into tasks ready to send (and download).
They took a really long time to be downloaded.

In the meantime, work seems to be trickling along fine. I've filled up my rigs with Rosetta again, hope this run lasts a while

For the moment. I doubt it'll last much longer tbh - perhaps an hour or two at most unless more get added to the queue

Yup, that's our lot for this batch from an hour or two back

Milan/Bergamo, Paris & London
And today Cardiff and back - a little less glamorous

In the meantime, maybe something like an extra 120k tasks? Hard to tell exactly.
Less than a day's worth, but still some left to grab for a few hours more

First batch of Beta work since mid-July.

Today we are releasing RFdiffusion3 as open-source software. This state-of-the-art AI model for biodesign is capable of generating new proteins that interact with any type of molecule commonly found inside living cells.
RFdiffusion3 delivers several key innovations:

Efficiency and performance: RFdiffusion3 offers ten-fold faster performance over RFdiffusion2 developed earlier this year. In computational benchmarks, it matched or outperformed prior tools on a wide range of protein-protein binding, protein-DNA binding, protein-small molecule binding, and enzyme design tasks.
Atom-level diffusion: The model treats individual atoms as the fundamental units being designed, applying a deep learning technique called diffusion to rapidly create new atomic arrangements. This produces intricate chemical interactions with unprecedented precision.
A unified foundation model: RFdiffusion3 consolidates design capabilities that previously existed only in scattered, specialized models. Whether designing symmetric, binding, or catalytic proteins, researchers can now use a single, general-purpose tool.