Bjørn Kwee

21 posts

Bjørn Kwee banner
Bjørn Kwee

Bjørn Kwee

@Bjornformatics

PhD student @Schumacher_Lab @NKI_nl | protein LLMs | Bioinformatics | Cancer Immunology

Amsterdam, Nederland Katılım Ocak 2021
517 Takip Edilen85 Takipçiler
Bjørn Kwee
Bjørn Kwee@Bjornformatics·
Incredibly excited and grateful for the opportunity to present at Antwerp TCR in 3 weeks. There’s no place more relevant to showcase what we have been up to. If you want to have a sneak preview; check out the thread of @MariusMessemak1 & our pre-print biorxiv.org/content/10.110…
Pieter Meysman@chevaliersf

#ATCR25 acceptance notifications just went out! We had 76 abstracts of very high quality, but could only accept 8! for short talks (so 10%). Huge thanks to our review committee for making though decisions. Really excited where the #TCR field is going based on the abstracts I saw!

English
1
3
9
778
Bjørn Kwee
Bjørn Kwee@Bjornformatics·
@iskander I would suggest looking at 3D, supplemental 3D and 4B. I am not entirely convinced that measuring CDR1 and CDR2 would solve the data quality problem, but happy to hear your thoughts.
English
0
0
1
58
alex rubinsteyn
alex rubinsteyn@iskander·
Going to read this and write a thread but definitely matches my strong suspicion from working with CDR3a/b annotations from IEDB/CEDAR/VDJdb -- though I've never been sure if matching to pMHC is wrong or just that CDR1 and CDR2 are too important to omit.
MariusMessemaker@MariusMessemak1

It's not the models, it's the data! We show that a substantial proportion of a widely used TCR-pMHC database does not functionally validate, causing underestimation of performance of TCR-pMHC prediction models.🧵 biorxiv.org/content/10.110…

English
3
1
12
1.5K
Bjørn Kwee
Bjørn Kwee@Bjornformatics·
@andimscience Thank you Andreas! Are you coming to TCR Antwerp this year? Would be great to catch up.
English
1
0
0
81
Andreas Tiffeau-Mayer
Andreas Tiffeau-Mayer@andimscience·
Very important work! We've been waiting for someone to do this. It has been clear for some time that a proportion of aggregated TCR-pMHC data might not be reliable. It will be interesting to reassess performance in light of this more robustly validated data.
Bjørn Kwee@Bjornformatics

It's not the models, it's the data! We show that a substantial proportion of a widely used TCR-pMHC database does not functionally validate, causing underestimation of performance of TCR-pMHC prediction models.

English
1
3
17
942
Bjørn Kwee
Bjørn Kwee@Bjornformatics·
@BingxuL Very interesting work Bingxu, congrats!
English
0
0
0
39
Bingxu Liu
Bingxu Liu@BingxuL·
We are all born with a genetic lottery. Millions of T cell receptors are what we have with a hope to defend all cancer and viruses. What if that's not enough? Hope our work can give an interesting answer to you. biorxiv.org/content/10.110…
English
9
45
236
27.6K
Bjørn Kwee retweetledi
Cancer Grand Challenges
Cancer Grand Challenges@CancerGrand·
Congratulations David Baker (@UW, member of #CancerGrandChallenges team MATCHMAKERS), who's won the 2024 Nobel Prize in Chemistry for his pioneering work building new kinds of proteins! 👏 More about MATCHMAKERS, funded by @CR_UK @theNCI @TheMarkFdn 👉 cancergrandchallenges.org/matchmakers
The Nobel Prize@NobelPrize

It's not easy to get hold of a newly awarded Nobel Prize laureate! We are happy that chemistry laureate David Baker took our call, as he turned down "100 calls" while he was on the phone with us. Hold tight for our upcoming interview with Baker.

English
2
8
18
3.6K
Bjørn Kwee retweetledi
Science and Innovation at Cancer Research UK
Science and Innovation at Cancer Research UK@CRUKresearch·
Ton Schumacher is closing #CancerHostTI24 discussing neoantigens and neoadjuvant therapy. With his research group he created technologies to dissect T cell responses in cancer and contributed to the development of adoptive T cell therapies and neoadjuvant cancer immunotherapy.
Science and Innovation at Cancer Research UK tweet media
English
1
4
19
1.7K
Bjørn Kwee
Bjørn Kwee@Bjornformatics·
@chevaliersf Thanks! I find the last point especially interesting. Would you say that your case (5000 epitopes with 100 TCRs) is preferred over, let's say, 50000 epitopes with 10 TCRs?
English
1
0
0
25
Pieter Meysman
Pieter Meysman@chevaliersf·
@Bjornformatics That being said, there are many other points to consider, such as HLA diversity, CD4/CD8 differences, etc
English
1
0
0
41
Bjørn Kwee
Bjørn Kwee@Bjornformatics·
@chevaliersf Very interesting! I watched the seminar, hoping that you would elaborate on these two numbers. Could you reference a paper or share the calculation you made to estimate these numbers? And how would the TCRs (ideally) be shared over the epitopes?
English
1
0
1
25
Pieter Meysman
Pieter Meysman@chevaliersf·
6. We are limited by data, not models, for unseen epitope prediction. 7. We need about 500 000 TCRs covering at least 5000 epitopes to get good unseen epitope performance. This is feasible in the next 2-3 years.
English
1
1
7
273
Bjørn Kwee retweetledi
Ton Schumacher Lab
Ton Schumacher Lab@Schumacher_lab·
Excited to share STAPLER, a language model to predict TCR – pMHC reactivity that outperforms prior models. And for ML aficionados: Description of a new data leakage problem inherent to a common negative data generation strategy. biorxiv.org/content/10.110… 🧵👇
English
2
34
107
27K
Bjørn Kwee
Bjørn Kwee@Bjornformatics·
Preprint alert 🚨🚨 Large language models can predict TCR - pMHC reactivity!🧬🔬 Proud to have worked on this project with @MariusMessemak1 and all other amazing co-authors.
Ton Schumacher Lab@Schumacher_lab

Excited to share STAPLER, a language model to predict TCR – pMHC reactivity that outperforms prior models. And for ML aficionados: Description of a new data leakage problem inherent to a common negative data generation strategy. biorxiv.org/content/10.110… 🧵👇

English
1
3
7
1.1K

Keşfet

@MariusMessemak1 @iskander @Paimaamu @andimscience @Schumacher_lab @BingxuL @UW @CR_UK