Ben Lai

AI Fellow @czbiohub Chicago| Prev PhD @TTIC_Connect

[Xinming Tu @TuXinming]

Also shouting out @zhou_jian who did a lot of pioneering work on sequence-to-function models. DeepSEA is one of the earliest papers in this field, and he continues to make great contributions to the space

[biohub @biohub]

New in @NatureBiotech from Aly Khan: a platform combining yeast display with fine-tuned protein language models to identify antigens driving autoimmune disease. In collaboration with Chris Garcia at @Stanford / @hhmi_science 🧵

[Ben Lai @blaiTTIC]

@sokrypton @DeryaTR_ I bet the demand of scientists for experimental structure determination and functional characterization increased instead of decreased after we “solved” the protein folding problem with AI.

[Ben Lai @blaiTTIC]

@keunwoochoi Exactly, most confusing analogy ever. Tried to look for where the “grad” is coming from and realize it’s imaginary 20 mins later.

[Ben Lai @blaiTTIC]

@DdelAlamo This has been the case for a couple of recent works, fine tuned pLMs are less effective in structure prediction tasks. Some of them don’t want to highlight this because they spent a lot of effort on their pLMs.

[Ben Lai @blaiTTIC]

Works well if your drug falls in the 20 ligands you are allowed to screen.

[Magdalena Skipper @Magda_Skipper]

Major AlphaFold upgrade offers boost for drug discovery https://t.co/bnj5wCEJdu

[bioRxiv Bioinfo @biorxiv_bioinfo]

Fast and accurate modeling and design of antibody-antigen complex using tFold https://t.co/UtlJKnHynb #biorxiv_bioinfo

[Ben Lai @blaiTTIC]

It was my pleasure to work with you and the team on this exciting hybrid protein engineering platform! Can't wait to see the potential applications that come out of this!

[Aerin Yang @aerin_yang]

Delighted to share a summary of my recent research paper on Deploying synthetic coevolution and machine learning to engineer protein-protein interactions! We
developed a new approach for coevolution of protein interfaces! (1/8)
https://t.co/tu9xkWyyjZ

[Horace He @cHHillee]

I suspect GPT-4's performance is influenced by data contamination, at least on Codeforces.

Of the easiest problems on Codeforces, it solved 10/10 pre-2021 problems and 0/10 recent problems.

This strongly points to contamination.

1/4

[Ben Lai @blaiTTIC]

@davis_yoshida ChatGPT also makes up nonexistent papers in a similar fashion. It’s kinda frustrating because I really wish those paper are real.

[Ben Lai @blaiTTIC]

@KevinKaichuang @alexijielu @amywang01 I was thinking 4n6t, guess I’m more of a fan for de novo folds.

[Ben Lai @blaiTTIC]

@sokrypton @structplantbio @buildmodels Am I the only one that feels cringe to see pLDDT ~ 60 models show up in PDB like this?

[Ben Lai @blaiTTIC]

@robin_andersson @Salvatore3Marco Code:https://t.co/mqaWNUuSra
Server:https://t.co/n3BK3kBODn

[Ben Lai @blaiTTIC]

@robin_andersson @Salvatore3Marco Nice Work! We have explored transfer learning for regulatory profiling in the context of the neural developmental process in https://t.co/1jWsGoKsop with a very similar framework named MetaChrom. It will be interesting to see the comparison of performance between these methods.

[Ben Lai @blaiTTIC]

@KevinKaichuang Thanks for mentioning our work, and stay tuned! We have more sequence optimization and modular design results coming.

[Ben Lai @blaiTTIC]

@takuma_yoneda Nice, I have some real wild prompts for you to try🤣🤣🤣

[bioRxiv Bioinfo @biorxiv_bioinfo]

End-to-End deep structure generative model for protein design https://t.co/oHtpQJoJ8k #biorxiv_bioinfo

[Hannes Stark @HannesStaerk]

The best mentor

[PLOS Comp Biol @PLOSCompBiol]

Read this new article, "Annotating functional effects of non-coding variants in neuropsychiatric cell types by deep transfer learning", here: https://t.co/L8Poj0ACg3

[juan @juanbuis]

it’s incredible how reading this feels like finally unlocking 100% of your brain

(via @christophepas)