Powers

Ph.D. Bioengineering @UOKnightCampus • ML for Cyclic Peptides @ParisaH_Lab & @SimonsFdn • Coding • Chemistry • Learn Languages (ES, IT)

[Calin Plesa @calinplesa]

New preprint! We carried out the largest complementation and inhibition study for diverse homologs in a single protein family (DHFR) to date. Highlights thread follows... 1/n 🧵

[Andrew White 🐦‍⬛ @andrewwhite01]

We had a lot to announce, but want to highlight we're building a PaperQA2-version of Wikipedia covering the human proteome. The 240 articles that were graded by experts as better than existing Wikipedia are already viewable - we're generating the rest over the next few weeks!

[Sergey Ovchinnikov @sokrypton]

Welcome back @ParisaH_Lab ! 🙂
https://t.co/hijRElSI7W

[Daniel Probst @[email protected] @skepteis]

Yes, Gzip can also classify molecules. You can find everything below.
If you want to help me tune it and publish it, let me know, my plate is full at the moment, but I couldn't resist quickly giving this a try.

#MachineLearning "#ArtificialIntelligence"
https://t.co/zrNvfiKMtt

[Kevin K. Yang 楊凱筌 @KevinKaichuang]

Break proteins into fragments, run DFT calculations, train a neural net to imitate DFT, and do ab initio simulations of protein dynamics

Tong Wang, Xinheng He, Mingyu Li, Yusong Wang, Zun Wang, Shaoning Li, Bin Shao, Tie-Yan Liu @MSFTResearch

https://t.co/282OojIbFn

[Kevin K. Yang 楊凱筌 @KevinKaichuang]

A tunable method for optimizing degenerate codon libraries that sample protein variants that are likely to have both high fitness and high diversity.

@jsunn_y Julie Ducharme @kadinaj @francescazfl @yisongyue @francesarnold

https://t.co/9Ss032KKFh

[Sergey Ovchinnikov @sokrypton]

AlphaFold inverted to hallucinate denovo proteins of up to 600 amino acids in length🤯

(animation below shows the designed protein docked into CryoEM density)

Exciting work with:
@chrisfrank662, @AKhoshouei, Yosta de Stigter, Dominik Schiewitz, @ShihaoFeng18, @hendrik_dietz

[Powers @ProteinPowers]

Found an oldie but a goodie paper this week. 1994 using Mixture Density Networks to predict multimodal distributions for a given prediction. Check it out if you want to get a bayesian machine learning perspective!
#BioE610

https://t.co/KasGr0SWyU

[Powers @ProteinPowers]

An interesting text-generation diffusion-based model to generate parts of a sentence at once and not token by token. The authors talk about the model being able to handle replacement, insertions, and deletions.
#BioE610

https://t.co/lljVVf6KYv

[Powers @ProteinPowers]

This excellent deep-learning paper can take unlabeled AA sequences and turn them into a statistical representation. Which is helpful for protein engineering tasks. #BioE610
https://t.co/aB0h0QXxgu

[Powers @ProteinPowers]

This is an interesting paper discussing how to integrate group invariant data into normally non-group invariant unsupervised deep learning. They use a novel function of encoding the information in a group invariant representation. #BioE610

https://t.co/EuiYGjIdkN

[Powers @ProteinPowers]

OSU was great! Thanks, Captain @David_Frey_ for getting us there safe!

[Gabriella Lindberg @LindbergLabs]

Thank you for hosting us today @OregonState, great to be able to share our classes in the field of #Bioengineering. @ParisaH_Lab, Adam Higgins, @GiersLab, @WillettLabs @meltelectrospin. @UOKnightCampus students got extra lucky, having our very own captain piloting our vehicle!

[Powers @ProteinPowers]

Cool new paper using the context of the pocket to generate reliable conformations for small molecules! Generative models are becoming more and more useful in a protein engineering context! #BioE610

https://t.co/gGFaL162mh

[Powers @ProteinPowers]

Diffusion models have been getting a lot of attention in regards to protein docking and binding site design. Really cool paper showing its applications. #BioE610

[Kevin K. Yang 楊凱筌 @KevinKaichuang]

Use a diffusion model to dock proteins and ligands and to design binding sites.

@ZhuoranQ @wn8_nie @ArashVahdat @tfmiller3 @AnimaAnandkumar

https://t.co/ttQ1PykX0R

[Powers @ProteinPowers]

This is a great paper on the use of reinforcement learning to design protein structure. Starting from the context of the shape and the designing the building blocks to fit it. Really cool use of Monte Carlo tree search Algo! #BioE610

[Shunzhi Wang @wang_shunzhi]

Excited to share our work on "top-down" reinforcement learning based approach to design proteins with exquisite structural control. The Monte Carlo tree search algorithm generate protein monomers and assemblies, guided by input geometric constraints. 1/4 https://t.co/OQvRqXSEbF

[Powers @ProteinPowers]

Excited to start classes next week with my cohort!

[Knight Campus @UOKnightCampus]

Whew, what a week! And what a great group of bioengineering Ph.D. students! That's a wrap for #ImpactWeek at the #KnightCampus. The bootcamp may be over, but the personal and professional growth will continue! https://t.co/FVXPmqNx3F

[Powers @ProteinPowers]

This is great!

[Christian Dallago @sacdallago]

I can finally blab about BioNemo OMG it was so difficult to keep this under wraps

https://t.co/a9AOfCUeHh

[David Frey @David_Frey_]

Having loads of fun already!

[Powers @ProteinPowers]

@shrutibalasa Set up a lot of print statements to figure out where my error is first 🤔