@rostlab

Predict protein structure and function using sequence, evolution & more! Operated by lab members.

 Munich & New York
rostlab.org
Joined November 2010
  • Tweets499
  • Following73
  • Followers733
229Media
Introducing ProtSpace: a tool for visualizing protein embeddings in 2D/3D! 🎨 biorxiv.org/content/10.1101/… Try it: 📦github.com/tsenoner/protspac… 🌐protspace.rostlab.org #Proteomics #DataViz
1
7
346
Very happy to introduce VespaG, for ultra-fast and accurate prediction of protein mutational effects. github.com/JSchlensok/VespaG. Big thanks to @LiniMuc, Julius Schlensok, and @MarinaAbakarova for the great work and effort. Exciting collab with @rostlab 1/5

GitHub - JSchlensok/VespaG: Expert-Guided Protein Language Models enable Accurate and Blazingly...

Expert-Guided Protein Language Models enable Accurate and Blazingly Fast Fitness Prediction - JSchlensok/VespaG

github.com
bioRxiv Bioinfo@biorxiv_bioinfo28 Apr 2024
VespaG: Expert-guided protein Language Models enable accurate and blazingly fast fitness prediction biorxiv.org/cgi/content/shor… #biorxiv_bioinfo
2
17
53
6,051
Our new bilingual protein language model (pLM), ProstT5, translates between protein sequence and structure. Besides producing more structure-aware embeddings that are better at remote homology detection than sequence-pLMs, its translation capability enables inverse folding.
1
28
1
121
30,874
1/3 🐍🧬 Exciting breakthrough in #SnakeToxins research published in #NatureCommunications! An international team led by Rostlab members unlocked the mysteries of three-finger toxins (3FTxs), from the venom of caenophidian snakes. 🔬#VenomResearch rdcu.be/djfyt
2
1
4
647
3/3 This study showcases the combined power of cutting-edge AI, traditional phylogenetics and manual annotations to unravel complex biological phenomena, such as evolution of protein function. Stay tuned for more! 📚🌟 #ScienceTwitter #Toxinology
1
277
2/3 🔍Immediate ancestor of these toxins? A non-secretory, non-toxic Ly6, unique to squamate reptiles. Mechanism? Exon duplication followed by mutation led to novel forms with a membrane anchoring domain lost. 🧩 #Evolution #ReptileScience
1
232
We (@lindner_seb, @HeinzingerM) introduce REXzyme, a new model for the generation of enzymes that catalyze user-defined chemical reactions. We'd like that the community can benefit as soon as possible from it, so we made it available in HuggingFace: huggingface.co/AI4PD/REXzyme.
6
53
2
208
25,855
Mar 03 - Oberseminar: This is only a test tinyurl.com/2jvkknwh
303
Did you know that 3-finger toxins in snake venoms evolved from a non-secretory ancestor? Our new preprint explores this gene family's evolution. Check it out here: biorxiv.org/cgi/content/shor… #geneevolution #proteinlanguagemodels #transmembranehelices
1
2
10
1,674
more replies
Contrary to previous hypotheses, our research suggests that 3-finger toxins in snake venoms evolved from a single-copy gene unique to Toxicofera reptiles, not monomeric neuromodulatory Ly6s like LYNXs or SLURPs. #snakevenom #geneevolution #proteinfamily
1
1
2
583
Synteny analyses, phylogenetics and pLMs reveal that 3-finger-toxins evolved from a transmembrane "pre-3FTx" ancestor that gained increased affinity for nAChRs and lost its transmembrane region. #snakevenom #geneevolution #machinelearning
1
3
514
Is ColabFold's MMseqs2-based MSA generation protocol suitable for mutational outcome prediction? Yes! Joint work by @MarinaAbakarova and Céline Marquet. 1/2
bioRxiv@biorxivpreprint15 Dec 2022
Alignment-based protein mutational landscape prediction: doing more with less biorxiv.org/cgi/content/shor… #bioRxiv
1
17
88
15,110
EMBER3D predicts protein structure in fractions of a second with high sensitivity to small changes in sequence. Protein Mutation Movies (PMMs) visually capture the effects of amino-acid substitutions. youtu.be/0tIN9n6cSHo via @YouTube
1
21
1
107
more replies
Trading accuracy for speed, EMBER3D runs orders of magnitude faster than competing models while maintaining reasonable performance (although nowhere close to Alphafold2).
1
2
6
EMBER3D can predict mutated structures of all single amino-acid variants in an average length protein in a matter of minutes. When correlated with deep mutational scanning experiments, EMBER3D predictions are much more sensitive to small changes in sequence than other systems.
1
2
8
SETH predicts nuances of disorder w/o alignments in milliseconds using pLMs and reaches (and sometimes outperforms) state-of-the-art methods. Manu: doi.org/10.1101/2022.06.23.4…
2
5
1
15
Update: SETH's disorder prediction correlates with AlphaFold2 pLDDT! If SETH predicts high disorder for a sequence you can assume (up to a 5% error) that the resulting AF2 structure will have low pLDDT, saving some compute/energy if you run structure prediction at proteome level.
1
2
3
15
Also: you can predict entire proteomes in minutes or hours (SwissProt ~7h), and the overall predicted disorder content draws a possible evolutionary path which fits expectation. What's also different than most disorder preds: not just binary, but richer CheZOD!
2
Since the # of "[α-ω](fold){0,1}[A-z]*" tools are getting out of hand, I tried to curate a lil list of the ones I'm aware of: github.com/sacdallago/foldin… Peeps w/ overview @sokrypton @thesteinegger @KevinKaichuang feel free to contribute if I missed stuff?
7
56
4
210
Load more