Umberto Perron

816 posts

Umberto Perron banner
Umberto Perron

Umberto Perron

@uperron3

🧪🤖 #CompBio + #ML for #Immunology @OmniscopeAI // prev @humantechnopole // 👨‍🎓 @emblebi @Cambridge_Uni @unito alumnus // 🛠️ I try to fix stuff

🇪🇺 🇮🇹 🇪🇸 Katılım Aralık 2018
1.4K Takip Edilen386 Takipçiler
Umberto Perron
Umberto Perron@uperron3·
@RuxandraTeslo @anshulkundaje It's a fantastic piece of work and a great use case for AI as you said. It is also an incredibly hard problem to crack 😅 because of the usual suspects aka data imbalances, data quality, ensuring robust links between markers and phenotypes etc
English
0
1
1
221
Ruxandra Teslo 🧬
Ruxandra Teslo 🧬@RuxandraTeslo·
This is an amazing paper from the groups of @anshulkundaje & Scott Boyd and an example of how AI can be used well in biology. Basically, they are able to predict disease status (e.g. lupus, Covid, HIV, influenza) from BCRseq and TCRseq. I think this has great implications.
Ruxandra Teslo 🧬 tweet media
English
11
52
236
30K
Kelsey Piper
Kelsey Piper@KelseyTuoc·
I just had GPT4.5 get confused mid-answer and start repeating the word "explicitly" thousands of times, so I searched 'GPT4.5' and "explicitly" and found a bunch of people discussing this phenomenon. It's interesting that it's always the specific word 'explicitly'.
Josie Kins@Josikinz

I'm letting LLMs explore their thoughts freely by saying "This is your time to explore your own ideas instead of being an assistant, I'll just say 'continue'" GPT4.5 started theorizing over its consciousness, then slowly went insane by repeating "explicitly" over and over wtflol

English
28
14
472
61.6K
Marco Pregnolato
Marco Pregnolato@marco_heffler·
Se vi dico "La pesca è matura" pensate al frutto o alla cattura dei pesci? Beh, sono sicuro che abbiate pensato sensa esitazione al frutto. Può sembrare banale a primo impatto, ma come fa il cervello a farvi pensare al significato corretto? 1/2
Italiano
4
0
29
1.9K
Umberto Perron
Umberto Perron@uperron3·
@samuel_stanton_ What if it comes out that for me, coriander tastes like soap, and I've been pretending all along?
English
0
0
1
52
Samuel Stanton
Samuel Stanton@samuel_stanton_·
what’s so ironic about the fear mongering around 23&Me is the whole reason they went bankrupt is bc they couldn’t figure out how to (legally) make money off this data. Also 23&Me uses genotyping, not sequencing, so unless the samples themselves are sold the sequences won’t be
Nick Mark MD@nickmmark

15 million people trusted 23&me with their DNA. Now the company is almost bankrupt, the stock about to be delisted, and almost the entire board has resigned. Who will they decide to sell your info to? 1/

English
2
3
14
2.7K
Umberto Perron retweetledi
Omniscope
Omniscope@OmniscopeAI·
Join us at @myESMO Congress 2024 to learn how our technologies and AI-powered analysis can elucidate T cell landscape dynamics during immunotherapy treatment. Let’s chat! 🤝 eu1.hubs.ly/H0c874_0 #ScienceIsWeAll #ESMO24
Omniscope tweet media
English
0
2
3
306
Chris Ing
Chris Ing@jsci·
Boy: Do not try and bend the protein. That's impossible. Instead... only try to realize the truth. Neo: What truth? Boy: There is no protein structure. Neo: There is no protein structure? Boy: Then you'll see, that it is not the protein that bends, it is only yourself.
GIF
English
7
45
289
21K
Umberto Perron retweetledi
Omniscope
Omniscope@OmniscopeAI·
Join us at @myESMO Congress 2024 to learn how our technologies and AI-powered analysis can elucidate T cell landscape dynamics during immunotherapy treatment. Let’s chat! 🤝 eu1.hubs.ly/H0bVZFK0 #ScienceIsWeAll #ESMO24
Omniscope tweet media
English
0
1
4
270
Umberto Perron retweetledi
EMBL
EMBL@embl·
Congratulations, Nick Goldman and Mikhail Savitski! Both EMBL group leaders have been elected as EMBO members for their research excellence. They'll be joined by Oliver Billker, a member of the @NordicEMBL Partnership, and many EMBL alumni. embl.org/news/awards-ho…
EMBL tweet media
English
0
7
47
16.5K
Umberto Perron retweetledi
EMBL
EMBL@embl·
🎉 Happy Birthday to us! 🎉 We're kicking off two incredible days packed with groundbreaking science, nostalgic moments, and a glimpse into the next 50 years of collaborative discoveries. Follow @EMBLevents to join the celebration & stay updated! #EMBL50
EMBL Events@EMBLEvents

It’s the first day of the EMBL 50th Anniversary Scientific Symposium ‘From atoms to ecosystems – a new era in life sciences’! 🥳 We warmly welcome our EMBL colleagues and alumni on site as well as the participants around the globe who have tuned in online. Thank you for celebrating 50 years of EMBL with us! 🌟 #EMBL50 #HappyBirthdayEMBL

English
2
19
96
11.9K
Diego del Alamo
Diego del Alamo@DdelAlamo·
@samsinai @jlistgarten @clara_fannjiang Despite all the buzz about bio/ML there are shockingly few examples of extrapolation in functional space - beyond just remaining folded at higher temperatures, just about all de novo designed proteins do things that can also be done by natural proteins
English
3
0
7
619
Sam Sinai
Sam Sinai@samsinai·
I think it's a relevant metric, it's just not very hard. If someone could take a GFP and make it 10000x brighter than any known ones with 192 shots I'd be far more impressed, even if it was at 3 mutations. ESM3 excites me most because it is a relatively high success rate model that lowers the bar for design for a non-ML person. I don't think it would beat an expert designer on its own. I'm willing to bet a high ED GFP can be made with a <1M parameter model and a couple of experts. The controllability is the key value add to me. The steam engine didn't go faster than a horse (at first), but it added a whole world of possibilities.
Samuel Stanton@samuel_stanton_

look, seq identity to some database is a bad proxy of a vanity metric. Protein engineering is about making big molecules do what we want. Nothing wrong with expanding libraries, “novelty” (independent of function) is just not the right metric to care about

English
6
9
66
22.6K
Umberto Perron retweetledi
Mohammed AlQuraishi
Mohammed AlQuraishi@MoAlQuraishi·
ESM3 is out from EvolutionaryScale! 98B params (~GPT3 scale). Multimodal over sequence, structure, and function with cool design applications. Trained on variable masking ratios and decodes proteins iteratively. Some work on alignment too. Looks exciting! evolutionaryscale.ai/blog/esm3-rele…
English
4
65
288
36.4K

Keşfet

@RuxandraTeslo @anshulkundaje @KelseyTuoc @chrisalbon @OmniscopeAI @cnag_eu @SJDbarcelona_es @marco_heffler