Share this page!

Enter URL or ID to Unroll

You can paste full URL like: https://x.com/threadreaderapp/status/1644127596119195649
or just the ID like: 1644127596119195649

How to get URL link on X (Twitter) App

  1. On the Twitter thread, click on or icon on the bottom
  2. Click again on or Share Via icon
  3. Click on Copy Link to Tweet
  4. Paste it above and click "Unroll Thread"!
  5. More info at Twitter Help
Pratyush Tiwary Profile picture
Pratyush Tiwary
@tiwarylab
Millard and Lee Alexander Professor in Chemical Physics at U Maryland. Molecules ∩ thermodynamics ∩ stat mech ∩ AI. Lab account but mostly dog + running tweets
Sep 8, 2023 • 8 tweets • 3 min read
Super excited to present latest preprint led by @bodhivani + grad student Akash - sampling DFG conformations in kinases with Boltzmann ranks using AlphaFold2RAVE

preprint:
codes:

This one warrants a detailed thread! (1/7) arxiv.org/abs/2309.03649
github.com/tiwarylab/kina…
Key idea: AlphaFold2 with reduced MSA as pioneered by @DdelAlamo can indeed sample alternate states of kinases. However most kinases look similar even with super low MSA - see here src, abl & our hero, DDR1 kinase. Thus AF2 sees most kinases as too similar! Now, why DDR1? (2/7) Image
Jan 2, 2023 • 5 tweets • 2 min read
I'm looking for sabbatical opportunities in industry for 2024!My ideal host would be a firm of any size involved in driving how AI &/or numerical modeling schemes are applied for betterment of human health&/or materials discovery.Details in thread,plz email/DM if interested (1/5) Details of what I'm looking for: 6 month or 1 year period engagement, starting any time in 2024 (as early as Jan 1 - it takes time to get things approved!) It could be in US, Europe or elsewhere. I am very open to the type of firm - in size and in the type of work (2/5)
Jul 16, 2021 • 6 tweets • 3 min read
New preprint on making best of replica exchange (REMD) by generative AI post-processing,specifically Denoising Diffusion Probabilistic Models (DDPM/DPM) developed by @pabbeel @jaschasd @SuryaGanguli et al. We show DDPM can directly sample joint p(x,T)(1/6) arxiv.org/abs/2107.07369 where x=any number of coordinates sampled across replicas. Key idea is to treat temperature T as fluctuating variable & not control parameter,by considering instantaneous kinetic energy of system. REMD sampling provides us a sparse snapshot from this high dimensional p(x,T) (2/6)