Yuan Liu

Physics and art make for a powerful combination.

pChem, developed by @Yang_J_lab and coworkers, provides a computational pipeline for scoring chemoproteomic probes on their efficiency, modification homogeneity, and residue selectivity. Free to read link at rdcu.be/cR8SA

PSP: Million-level Protein Sequence Dataset for Protein Structure Prediction abs: arxiv.org/abs/2206.12240 dataset consists of 570k true structure sequences (10TB) and 745k complementary distillation sequences (15TB)

Protein structure can be predicted from a single sequence alone with high accuracy. @HelixonBio team have developed OmegaFold, achieving performance similar to RF and AF2's MSA versions. Only a single sequence is given as input. 1/5

Supper fast! Try MEQKPGTLMVYVVVGYNTDNTVDVVGGAQYAVSPYLFLDVGYGWNNSSLNFLEVGGGVSYKVSPDLEPYVKAGFEYNTDNTIKPTAGAGALYRVSPNLALMVEYGWNNSSLQKVAIGIAYKVKD with recycle = 12

@sokrypton looks like a knot

Going a little deeper into the rabbit hole. 🐇😅 I put together a "partial hallucination" example using af_design. That starts from the secondary structure (ss) segments of 2MR5 and hallucinates a new protein with ss order of 7BQD. (1/2)

Does anyone have a suggested metric for quantifying the similarity of folds? I was looking at TMalign, but this can be misleading. For example, these 2 structures have TMscore of 0.66, yet different folds. Left:2MR5, Right: novelfold 7BQD. (🌈-colored where architectures overlap)

@sokrypton is there any bug?

Another example (seems to work a little too good, time to investigate if there is any bug... leading to any leaking of information) 😅

Minor update to our decoy ranking manuscript. We (@jamesproney) add a plot showing that besides simply being able to rank decoys, AlphaFold can also be used to refine decoys: biorxiv.org/content/10.110… (1/5)

@awakenkunkun @kobnaoya 你的意思是pose中的所有信息？

@kobnaoya silent文件可以储存所有pose中的信息

結果を出力する時、dumpしているけれど、score fileも出させたい場合はJob Distributorとかを使えばいいのだろうか？

@ruanyf 不妨

昨晚做了一个梦，梦见奥克米戎病毒又发生了变异，传染性增强了10倍，但是感染者100%永久无症状。 醒了以后一想，觉得这个病毒太恶毒了，你让这边怎么防呢。

For those interested in activity-based protein profiling (ABPP), #FragPipe 17 comes with 3 workflows: classic isoTop-ABPP, SLC-ABPP (TMT16, DBIA tag nature.com/articles/s4158…), and isoDTB (chemrxiv.org/engage/chemrxi…). They can be easily customized. Let us know what else we should add.

@UCDProteomics We are planning to create a #FragPipe "workflows repo" (including shared by others, e.g. from published papers). In addition to the more common ones we already include. So click 'Save workflow" and share with us, if you found settings that worked better than our default TMT-Ubiq.

Our recent presentation on inverting structure prediction models for design is now public, for those that missed it! 😁 github.com/sokrypton/Cola… Features @jueseph and new analysis from @JustasDauparas @KotaroTsuboyama @grocklin @minkbaek @sid_thesci_kid and others @UWproteindesign

@sokrypton unbelievable

I tried hallucinating mini binders for a few examples from Amy Keating. Seems to work a little too good... not sure if this is "binder design" or "binder-site prediction/recapitulation" 😅 (green is the target, cyan is the binder).

Adding support for binder hallucination if anyone wants to try! (Code is very experimental, not intended for practical use... only use for art/science) 😀 colab.research.google.com/github/sokrypt…

Just as convincing images of cats can be generated using artificial intelligence, new proteins can now be made using similar tools. In a new report in @Nature, we describe a neural network approach for “hallucinating” proteins with new, stable structures. nature.com/articles/s4158…

Huge thanks to our many users (fast growing userbase!) and the feedback we get via github.com/Nesvilab/FragP…. We just released #FragPipe 17.1 with minor fixes, including in MSBooster and in the DIA workflows, that affected a small number of users. If you are one of them, upgrade.

Graph neural networks are rapidly advancing progress in ML for complex graph data applications. In this week’s newsletter, we show you recent uses of GNNs ranging from protein interface prediction to collaborative filtering! Read on below: paperswithcode.com/newsletter/19/

@fcyucn Is opensearch also available for DIA?

See you in 30 minutes #ASMS2021 #ASMS

@sokrypton @GoogleColab 😡

@GoogleColab Colab Pro+ seems to be broken. I upgraded last night hoping to increase runtime for a project... And the single Job was terminated after ~8 hours... now no more GPUs are available. 😭 (I had more resources w Pro... Now gotta wait till end of month to downgrade)

Impatient? Wanna see models as the notebook is running? 😎 See ColabFold AlphaFold2_advanced. Also, you can now use '/' to specify chain breaks. For example: A/C will modeled as A and C (could be used to trim disordered regions, or specify 2+ proteins) github.com/sokrypton/Cola…

@sokrypton make a Mobius circle

InfinityFold 🤓 (doing memory benchmarks, single-sequence, 1300 poly-A sequence)