Surge Biswas

@SurgeBiswas

Interested in what machine learning can do for protein engineering & sports. lab . PhD candidate . Dog dad.

Vrijeme pridruživanja: rujan 2013.

Tweetovi

Blokirali ste korisnika/cu @SurgeBiswas

Jeste li sigurni da želite vidjeti te tweetove? Time nećete deblokirati korisnika/cu @SurgeBiswas

  1. Prikvačeni tweet
    24. sij

    This has been a really fun one! challenging the wisdom that one needs a lot of data to do ML guided biodesign. In fact, if you have to choose, think about getting HQ small-N data aligned w your endpoint, vs HT proxy data. low-N ML now helps you succeed w the former!

    Prikaži ovu nit
    Poništi
  2. 30. sij

    🤯Using this to calculate tips from now on

    Poništi
  3. proslijedio/la je Tweet
    29. sij

    Cool new result in protein design. Multitask from large data to small - transfer learning from a general protein representation to one tailored to a particular function of interest.

    Poništi
  4. proslijedio/la je Tweet
    28. sij

    We're hiring a bioinformatician to help us tackle a number of projects involving large scale sequence design, optimization, and analysis:

    Poništi
  5. proslijedio/la je Tweet
    28. sij

    In DC for , so much cool work here in rapid countermeasure response for new pathogens! If you are working on this, I want to talk - dms open.

    Poništi
  6. proslijedio/la je Tweet
    27. sij

    Great work on using ML for protein engineering. Key insights: - fine-tuning global LM using local landscape helps - pretrained model can predict epistatic effects from single mutants

    Poništi
  7. proslijedio/la je Tweet
    25. sij
    Odgovor korisniku/ci

    This was fantastic! The PC1 correlation is actually just amazing.

    Poništi
  8. proslijedio/la je Tweet
    24. sij

    Interesting potential for ML-guided protein engineering: might be able to systematically optimize proteins after just a few choice measurements. Check out this new preprint from and team.

    Poništi
  9. proslijedio/la je Tweet
    24. sij

    won't stop me from working on ways to get big-N, but sometimes small-N is hard enough

    Poništi
  10. proslijedio/la je Tweet
    25. sij

    Super cool work from , , , , and ! Efficiently identifying mutant proteins with higher activity using ML, i.e., eUniRep. Congrats!!!

    Poništi
  11. proslijedio/la je Tweet
    24. sij

    Had an early look at this work and it’s really impressive stuff! Demonstrates the remarkable power of semi-supervised learning in very low N contexts.

    Poništi
  12. proslijedio/la je Tweet
    24. sij

    Our new paper! My favorite bits were: - Discovering WAY more functional proteins out there (1000+!!) then previously explored by evolution OR decades of engineering - Connecting engineering and tech-translation failures with Goodhart's law - 's first thread :P

    Poništi
  13. 24. sij

    Hope you enjoy the read! We haven't submitted this anywhere yet so feedback and venue/journal suggestions welcome! DMs open

    Prikaži ovu nit
    Poništi
  14. 24. sij

    As always, inspiring to work with and learn from and , and have the mentorship of and :)

    Prikaži ovu nit
    Poništi
  15. 24. sij

    For the ML folks reading this, we're excited by the extreme data-efficiency we saw here, and for forward design of a complicated natural object no less. Another feather in the cap of unsupervised/semi-supervised learning, no doubt!

    Prikaži ovu nit
    Poništi
  16. 19. sij
    Poništi
  17. 17. sij

    11/10 opportunity to work for a great company, , with an insanely fun and smart person!

    Poništi
  18. proslijedio/la je Tweet
    16. sij

    It's time to replace λBeta in MAGE, DIvERGE and ssDNA recombineering! By analyzing 100s of proteins, we have identified 2 that allow E.coli oligo-recombineering with up to 50% efficiency, increase multiplexability + allow MAGE in P. aeruginosa – mjesto: Harvard Medical School New Research Building

    , , i još njih 6
    Prikaži ovu nit
    Poništi
  19. proslijedio/la je Tweet
    16. sij

    Very happy to release publicly the RedBeta-killer, CspRecT. This protein enables 50% editing efficiency with single-stranded oligos in E. coli recombineering! Performance improves even more for multiplex applications like MAGE and DiVERGE.

    Prikaži ovu nit
    Poništi
  20. 14. sij

    A short and sweet ;) read. So little we understand about glycobio but a major limiting factor for therapeutic dev. Also gets relatively little academic attention (vs DNA/RNA/proteins), but prob lots of siloed industry knowledge Tools like these democratize their understanding!

    Poništi

Čini se da učitavanje traje već neko vrijeme.

Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.

    Možda bi vam se svidjelo i ovo:

    ·