Patrick Bryant @Patrick18287926 Twitter profile

Last Seen Profiles

@koiacats

@richardbherman

@aljntlaljn97198

@alexbchen

@eretsym57

@LeStyleParfait

@VanKonRab

@Maykuudere

@Penda_chui

@HashuruTV

@REVMAXXING

@La0Oq0T1I428ZNp

@420joke

@sslions0205

@AndrewPlaca

@MatiDapatCuki

@Couplemudajogj1

@JOuellette19

@Pwaterpolo

@TheRootOf_je

@houseofdelaa

@helydiaa

@John_Zhang_2010

@Xbox_KR

@Alojzypopiolek2

@MikeEconomo

@mussett8

@perwinc

@katieincolumbus

@LittleSoftShop

@stw_pdg

@ernestomen81

@K3v1n_du_71

@VAAlvarez_NIH

@RhyleighZa53824

@FinalKhubsun

Patrick Bryant

@Patrick18287926

8 months

Structure prediction of protein-ligand complexes from sequence information. State of the art. Data available+Code available (including a Colab).

6

138

561

Patrick Bryant

@Patrick18287926

21 days

I am hiring a PhD student in AI for structure prediction and design. The position is fully funded for 5 years (salary, health insurance, benefits) and will take place at Stockholm University/SciLifeLab.

3

129

270

Patrick Bryant

@Patrick18287926

9 months

Hi! I am starting a group at Stockholm University in AI for protein applications. Currently hiring a PhD student: and a postdoc: . Come work with me in beautiful Stockholm!

12

81

237

Patrick Bryant

@Patrick18287926

1 month

Happy to see Umol published: If you want to predict the structure of protein-ligand complexes and avoid being sued by Google or Nvidia - we are here for you. Great work with @AtharvaKelkar Andrea Guljas @CecClementi and @FrankNoeBerlin

Structure prediction of protein-ligand complexes from sequence information with Umol

Nature Communications - Here the authors report the AI system Umol that predicts flexible all-atom structures of protein-ligand complexes from sequence information, advancing AI-driven drug...

www.nature.com

1

55

226

Patrick Bryant

@Patrick18287926

9 months

We trained a new network for protein structure prediction on a conformational split of the PDB to generate alternative conformations. 52% (81) of the nonredundant protein conformations evaluated are predicted with high accuracy (TM-score>0.8). Try it out:

3

46

165

Patrick Bryant

@Patrick18287926

12 days

We designed linear and cyclic peptide binders only from a protein target sequence. Different lengths; 46% had Kd in the micromolar range from a single sequence selection.

Design of linear and cyclic peptide binders of different lengths only from a protein target sequence

Structure prediction technology has revolutionised the field of protein design, but key questions such as how to design new functions remain. Many proteins exert their functions through interactions...

www.biorxiv.org

8

32

172

Patrick Bryant

@Patrick18287926

1 year

Check out our new paper where we improve protein complex prediction by performing gradient descent through the AF-multimer network. We effectively denoise the MSA profile, similar to how a blurry image would be sharpened.

3

31

145

Patrick Bryant

@Patrick18287926

8 months

Update: Umol can predict affinity, even though we did not train for this(!). Here are affinity values vs ligand plDDT on a held-out test set. This is not in the preprint () - more details will come in the publication.

4

19

126

Patrick Bryant

@Patrick18287926

8 days

This is what it looks like. Only the receptor sequence (green) is input and EvoBind2 finds a binding spot and creates a binder (blue) simultaneously. Code:

Patrick Bryant

@Patrick18287926

12 days

We designed linear and cyclic peptide binders only from a protein target sequence. Different lengths; 46% had Kd in the micromolar range from a single sequence selection.

8

32

172

2

18

106

Patrick Bryant

@Patrick18287926

2 months

Come to Sweden instead. We pay you a reasonable salary, universal health care, 40h work week, 30 days vacation. Ranked 2nd in innovation worldwide

nature

@Nature

2 months

Graduate students are relying on donated and discounted food in the struggle to make ends meet.

11

66

161

4

0

45

Patrick Bryant

@Patrick18287926

3 months

I am hiring a PhD student in AI for drug development: Come join us in beautiful Stockholm! More info about the group:

3

20

40

Patrick Bryant

@Patrick18287926

8 months

Had a great time in Copenhagen! Fantastic hosts!

Frederik Post🇺🇦

@Freddyomics

8 months

Mind-blowing talk by @Patrick18287926 at the @cphbiosciphd Copenhagen Bioscience Snapshot about AI tools for the prediction of protein interactions! It was a pleasure for @VKleinSousa and me to host you. Try out his fast and state-of-the-art tools!

0

7

40

0

1

33

Patrick Bryant

@Patrick18287926

9 months

It is hard to tell what the performance here really is. They report 42% <2Å l-RMSD on PostBusters (), but train on all of PDB? DiffDock has 1% considering data-overlaps and chemical validity. Please mind your data.

Sergey Ovchinnikov 🇺🇦

@sokrypton

9 months

RoseTTAFold updated to be All-Atom... biological assemblies containing proteins, nucleic acids, small molecules, metals, and covalent modifications ... and diffusion🤯

8

250

886

3

5

31

Patrick Bryant

@Patrick18287926

9 months

Hi, don't miss out on this nice postdoc opportunity in protein design! Please note that you must apply here (and not email me): …

Patrick Bryant

@Patrick18287926

9 months

Hi! I am starting a group at Stockholm University in AI for protein applications. Currently hiring a PhD student: and a postdoc: . Come work with me in beautiful Stockholm!

12

81

237

1

6

22

Patrick Bryant

@Patrick18287926

8 months

Last day to apply. Come join me!

Patrick Bryant

@Patrick18287926

9 months

Hi! I am starting a group at Stockholm University in AI for protein applications. Currently hiring a PhD student: and a postdoc: . Come work with me in beautiful Stockholm!

12

81

237

0

3

20

Patrick Bryant

@Patrick18287926

2 years

This is now published including a substantial comparison with Alphafold-multimer where we highlight the success of MolPC for large protein complexes of different symmetry

Predicting the structure of large protein complexes using AlphaFold and Monte Carlo tree search

Nature Communications - The accuracy of AlphaFold decreases with the number of protein chains and the available GPU memory limits the size of protein complexes that can be predicted. Here, the...

www.nature.com

Arne Elofsson @arneelof.bsky.social

@arneelof

2 years

My extremely talented student Patrick has taken #alphafold one step further and shows that we are entering the era of automatic predictions of the structure of large complexes

8

111

491

1

3

20

Patrick Bryant

@Patrick18287926

12 days

First paper from my lab. Big thanks to my team, especially Qiuzhen Li! Code:

GitHub - patrickbryant1/EvoBind: In silico directed evolution of peptide binders with AlphaFold

In silico directed evolution of peptide binders with AlphaFold - patrickbryant1/EvoBind

github.com

0

2

17

Patrick Bryant

@Patrick18287926

8 months

Work with @AtharvaKelkar , Andrea Guljas, @CecClementi and @FrankNoeBerlin

2

1

12

Patrick Bryant

@Patrick18287926

2 months

@pushmeet Great to hear - please also wait to release the print in nature to maintain the scientific practice. I think we still have a long way to go in this space and industry-academia interplay is essential

1

0

10

Patrick Bryant

@Patrick18287926

2 months

AlphaFold3: will be available as a web server. The protein-ligand lDDT displays how far we have to go until these models really become useful. We are missing some crucial piece of information:

1

0

6

Patrick Bryant

@Patrick18287926

10 months

@sokrypton I can't read the paper - no Science access (I guess they couldn't afford the open-source fee). The results are underwhelming though and I don't see the point of predicting 71 million missense variants when the SpearmanR is 0.5 to ddG (I presume)

2

0

7

Patrick Bryant

@Patrick18287926

9 months

We went through the PDB and partitioned all single-chain structures into structural clusters. Then, we train on one partition and evaluate on the other. Result: learn to relate one MSA to one conformation and evaluate on all proteins with substantial conformational changes.

0

7

Patrick Bryant

@Patrick18287926

2 months

@pedrobeltrao Yeah, this will be an issue - not knowing the input info makes benchmarking impossible

0

4

Patrick Bryant

@Patrick18287926

9 months

Here is more about my previous research:

Research

Dedicated scientist.

patrickbryant1.github.io

Patrick Bryant

@Patrick18287926

9 months

Hi! I am starting a group at Stockholm University in AI for protein applications. Currently hiring a PhD student: and a postdoc: . Come work with me in beautiful Stockholm!

12

81

237

1

0

6

Patrick Bryant

@Patrick18287926

5 months

@thesteinegger @iscb @SoedingL @StevenSalzberg1 @milot_mirdita Congratulations 🎉 very well deserved I think. The tools you develop enable so much research!

1

0

5

Patrick Bryant

@Patrick18287926

3 years

@NobelPrize The correlation between telomere length and ageing is around 0.4 in humans. This statement makes it look like you actually know why we age/what ageing is. Please refer to the sources of this statement. Finding an enzyme that extends DNA ends hardly explains ageing...

1

0

5

Patrick Bryant

@Patrick18287926

3 months

The position is fully funded (salary, insurance etc) for 5 years. You don't need to know Swedish - python is enough.

0

1

4

Patrick Bryant

@Patrick18287926

2 months

@sokrypton Run and submit it to CASP as OvchinnikovFold? 😅

1

0

5

Patrick Bryant

@Patrick18287926

10 months

@arneelof Yeah, I think the ddG data is the key here and there the performance is low. I think what is being learned is evolutionary likelihoods and not the actual effects.

0

1

4

Patrick Bryant

@Patrick18287926

9 months

The work will take place at SciLifeLab. Here is more about me, scilife and the funding agency (DDLS):

New DDLS Fellow: Patrick Bryant

Get to know our latest SciLifeLab & Wallenberg National Program for Data-Driven Life Science (DDLS) Fellow, Patrick Bryant (Stockholm University), in this week’s Q&A-style article. Patrick will be...

www.scilifelab.se

0

3

Patrick Bryant

@Patrick18287926

1 year

@judith_bernett @itisalist @dbblumenthal Great to see that someone shows what many have assumed: 99% PPI accuracy from single sequence is unreasonable...Reviewers have to be better at checking data.

0

3

Patrick Bryant

@Patrick18287926

1 year

We demonstrate the performance on seven difficult targets from CASP15 and increase the average MMscore to 0.76 compared to 0.63 with AF-multimer(AFM). Available here:

0

3

Patrick Bryant

@Patrick18287926

9 months

Paper:

Structure prediction of alternative protein conformations

Proteins are dynamic molecules whose movements result in different conformations with different functions. Neural networks such as AlphaFold2 can predict the structure of single-chain proteins in the...

www.biorxiv.org

1

0

3

Patrick Bryant

@Patrick18287926

9 days

Two weeks left to apply! Come join us and design new molecules and develop new AI tools!

Patrick Bryant

@Patrick18287926

21 days

I am hiring a PhD student in AI for structure prediction and design. The position is fully funded for 5 years (salary, health insurance, benefits) and will take place at Stockholm University/SciLifeLab.

3

129

270

0

1

4

Patrick Bryant

@Patrick18287926

7 months

@oda_sci Looks very much like EvoBind (, ) and EvoPlay () , almost named the same, but I see that you didn't cite these. Why?

Self-play reinforcement learning guides protein engineering

Nature Machine Intelligence - There are currently promising developments in deep learning for protein design, with applications in drug discovery and synthetic biology. For more efficient...

www.nature.com

0

3

Patrick Bryant

@Patrick18287926

1 year

@amine_ketata Doesn't it say that HDock is better at 2Å? Or do you consider your oracle with perfect selection here? Btw, did you try to run the Hdock scoring function on your generated poses? That scoring function looks very good for many tasks

3

0

3

Patrick Bryant

@Patrick18287926

9 months

Neural networks such as AlphaFold2 see almost all conformations in the PDB during training. Therefore, it is not possible to assess whether alternative protein conformations can be predicted or if these are reproduced from memory.

1

0

3

Patrick Bryant

@Patrick18287926

1 year

Our protocol, AFProfile, provides a way to direct predictions towards a defined target function guided by the MSA. We expect gradient descent over the MSA to be useful for different tasks, such as generating alternative conformations.

0

3

Patrick Bryant

@Patrick18287926

1 year

@arneelof x2

0

3

Patrick Bryant

@Patrick18287926

2 years

@arneelof We mutate the binding residues of peptides and show that AF can distinguish between native and mutated peptide binders using only the plDDT score:

0

3

Patrick Bryant

@Patrick18287926

2 months

@AnimaAnandkumar How do you know? I think there should be very little difference btw any of these tools on new (low homology) targets. None work very well.

0

3

Patrick Bryant

@Patrick18287926

8 months

Andrea Guljas went through the PDB and extracted affinity values for a large number of complexes - great effort!

1

0

2

Patrick Bryant

@Patrick18287926

2 years

@kobnaoya @thesteinegger There is a bash script from PDB to do this if you just supply your hits:

Batch Downloads with Shell Script

As a member of the wwPDB, the RCSB PDB curates and annotates PDB data according to agreed upon standards. The RCSB PDB also provides a variety of tools and resources. Users can perform simple and...

www.rcsb.org

1

0

2

Patrick Bryant

@Patrick18287926

2 years

Check out my latest work on structure prediction of protein-protein interactions using AlphaFold2 published today in Nature communications:

Improved prediction of protein-protein interactions using AlphaFold2

Nature Communications - Predicting the structure of protein complexes is extremely difficult. Here, authors apply AlphaFold2 with optimized multiple sequence alignments to model complexes of...

www.nature.com

0

1

2

Patrick Bryant

@Patrick18287926

1 year

@sokrypton I have noticed similar issues with training and I think this is by design. If you delete the references to the function you are calling (e.g. del predict) after you call it this releases the cache. You can also do .clear_cache() depending on how things are vmapped/pmapped.

1

0

2

Patrick Bryant

@Patrick18287926

2 months

@jankosinski @br_lichtenstein Relax now and have a look at the training metrics. It is not possible to predict these things - and for good reasons.

0

2

Patrick Bryant

@Patrick18287926

7 months

@pedrobeltrao @sokrypton There is even published work building on EvoBind (EvoPlay, ) which is great 👍 Difference is they cite us. Maybe it's just me, but I think previous work should be cited. They should also cite EvoPlay.

Self-play reinforcement learning guides protein engineering

Nature Machine Intelligence - There are currently promising developments in deep learning for protein design, with applications in drug discovery and synthetic biology. For more efficient...

www.nature.com

1

0

2

Patrick Bryant

@Patrick18287926

2 years

@sokrypton Nice, when is this tutorial/is it possible to join?

2

0

2

Patrick Bryant

@Patrick18287926

9 months

@WillMcCorki1 Thanks, maybe mention this in the main text? The issue here is the overlap with PoseBusters. It would be nice to see the performance on proteins with <20% seqid to your dataset compared to the other methods, aka performance on unseen targets.

1

0

1

Patrick Bryant

@Patrick18287926

1 year

@y_bromberg Great to see these types of studies 👏. As expected, function prediction is more like annotation. Perhaps data partitions should not be based on date cutoffs for evaluation (CAFA)...

1

0

2

Patrick Bryant

@Patrick18287926

9 months

Swedish is not required (there seems to be some confusion about this). We do speak English in Sweden.

Patrick Bryant

@Patrick18287926

9 months

Hi! I am starting a group at Stockholm University in AI for protein applications. Currently hiring a PhD student: and a postdoc: . Come work with me in beautiful Stockholm!

12

81

237

0

2

Patrick Bryant

@Patrick18287926

8 months

@jchodera Great! Thanks, we definitely will

0

1

Patrick Bryant

@Patrick18287926

1 year

@mpi_nat @milot_mirdita @SNUnow @SoedingL It doesn't compare sequences, it compares an alphabet of structure representations. It looks like a description of BLAST now?

1

0

1

Patrick Bryant

@Patrick18287926

2 months

@FurmanLab What templates were used here? (How do you know the native structures are not input?)

1

0

1

Patrick Bryant

@Patrick18287926

2 years

@sokrypton Update: we have now added the possibility to upload your own receptor structures and to specify a starting sequence to optimise further. This also makes it possible to only predict by setting NITER=1. If you have a request, contact us. 😊

0

1

Patrick Bryant

@Patrick18287926

1 year

@arneelof @amine_ketata Hdock seems to be the best for rigid docking. It is also very fast. Quite impressive since the scoring function was trained on relatively few structures.

0

1

Patrick Bryant

@Patrick18287926

2 years

@LiZiyao11 You need to know the symmetry for this right?

1

0

1

Patrick Bryant

@Patrick18287926

2 years

@JameAbduljalil @sokrypton @arneelof We didn't. If this is what you have available in the colab session the largest trimer can't be bigger than 1000 residues. You could try it in dimer mode if this doesn't fit for you.

0

Patrick Bryant

@Patrick18287926

1 year

@sokrypton Nice 👍 perhaps this can help to resolve really large trees that are too big to bootstrap efficiently? May be hard to check though.

0

1

Patrick Bryant

@Patrick18287926

9 months

@nickpolizzi_ @sokrypton Important to also extract the available seqres from the files, otherwise you will end up with clusters/mappings that do not exist (don't trust the seqres from pdb)

1

0

1

Patrick Bryant

@Patrick18287926

2 years

@JameAbduljalil @sokrypton @arneelof There is no upper limit. This will depend on your available GPU memory. Currently, approximately 3000 residues fit on 40Gb which will be limited by the size of the largest trimeric subcomponent.

2

0

Patrick Bryant

@Patrick18287926

2 years

@arneelof AF adapts the receptor interface structure to the binders during optimisation. An example for PDB ID 2cnz is shown, where residue 12 (orange) in the receptor (grey, green) changes orientation to interact with the peptide residues (blue, magenta).

0

1

Patrick Bryant

@Patrick18287926

7 months

@eriklindahl @sokrypton @SamTipps I agree and hope this can be cleared up

1

0

Patrick Bryant

@Patrick18287926

7 months

@pedrobeltrao @sokrypton I think you are confusing the years. Theirs is from May 2023 () mine from July 2022 ()

In silico evolution of protein binders with deep learning models for structure prediction and...

There has been considerable progress in the development of computational methods for designing protein-protein interactions, but engineering high-affinity binders without extensive screening and...

www.biorxiv.org

1

0

1

Patrick Bryant

@Patrick18287926

7 months

@Deconstructdog Sounds like you don't read much then. Not surprised you haven't heard about a paper before today that came out today.

1

0

1

Patrick Bryant

@Patrick18287926

8 months

@jordanberg0 Thanks! Yes, I think so. If you input paired MSAs, but we have not tried. In the next version it will (and many other things as well)

1

0

1

Patrick Bryant

@Patrick18287926

2 years

@arneelof Check out the Colab notebook: (inspired by ColabFold @sokrypton @milot_mirdita @thesteinegger )

Google Colab Notebook

Run, share, and edit Python notebooks

colab.research.google.com

1

0

1

Patrick Bryant

@Patrick18287926

2 years

@sokrypton Very interesting. Maybe it is best to use AF, OF, OmegaFold and ESMFold? Will be interesting to see how one can reduce the adversarial effect through a joint score. Are we then back at ensemble methods?😅

0

1

Patrick Bryant

@Patrick18287926

9 months

@arneelof This is not the same problem though. By that logic, everyone can continue to evaluate on their training data. Sad that this is so widespread that it is not considered problematic.

2

0

1

Patrick Bryant

@Patrick18287926

7 months

@Lauren_L_Porter I think this is true. Note that ColabFold is not trained in any way. ColabFold is an online version of AlphaFold. What you are really discussing is AlphaFold and overfitting to certain sequences.

0

Patrick Bryant

@Patrick18287926

9 months

@sokrypton Completely agree with this. Not sure this is what they are saying though.

1

0

1

Patrick Bryant

@Patrick18287926

9 months

@DewaDNMAdyaksa Yes, all you need is the sequence. There are instructions here:

GitHub - patrickbryant1/Cfold: Structure prediction of alternative protein conformations

Structure prediction of alternative protein conformations - patrickbryant1/Cfold

github.com

0

1

Patrick Bryant

@Patrick18287926

10 months

@arneelof I don't think this technology is accurate enough to be useful. Predicting 71 million mutations may do more harm than good imo.

AlphaFold tool pinpoints protein mutations that cause disease

Nature - Researchers have adapted the AI network to search for genetic changes linked to ill health.

www.nature.com

1

0

1

Patrick Bryant

@Patrick18287926

9 days

@owl_poster Also like the idea, but 58% with <native activity will not be very useful. I would be surprised if a different fold was adopted at 58%. The active site is really what matters here

0

1

Patrick Bryant

@Patrick18287926

2 years

@ThatMolBioGuy @arneelof Read the preprint and find out (hint: coevolution)

2

0

1

Patrick Bryant

@Patrick18287926

4 months

@tiwarylab @HannesStaerk @GrantRotskoff @UWproteindesign @loukasa_tweet @KevinKaichuang @fra_grisoni @thesteinegger @marwinsegler @Blendenfleck @open_fold @nickpolizzi_ @inve_michele @AnimaAnandkumar @VStimper @jchodera @HeinzingerM @simonduerr @__kraemer__ @igashov I think the problem here is the test set. If you train on almost all conformations in the pdb (via transfer learning) you have very little left to test on. The "conformations" also have to be different, the states above are very difficult to distinguish.

1

0

1

Patrick Bryant

@Patrick18287926

2 years

@BrianHie Nice work! I would perhaps remove the conclusion regarding the correlation analysis in Figure 3B, this result I think is due to the large separation btw values and very few data points.

1

0

1

Patrick Bryant

@Patrick18287926

8 months

@KevinKaichuang @AtharvaKelkar @CecClementi @FrankNoeBerlin The wrong paper is linked here(?)

0

1

Patrick Bryant

@Patrick18287926

2 years

@ylecun You are the first person I hear agree with me on the power of gradient descent. Everyone I have ever talked to seems to think it is not such an important concept in ML. To me, the idea of gradient descent is what is important.

0

1

Patrick Bryant

@Patrick18287926

3 months

@Nature This is terrible.

0

1

Patrick Bryant

@Patrick18287926

9 months

The PhD position comes with full funding for 4 years, salary, insurance, pension package, health benefits and help with student housing (the same package as for all employees).

1

2

1

Patrick Bryant

@Patrick18287926

2 months

@FrankNoeBerlin @olexandr Maybe it can't be done?

0

1

Patrick Bryant

@Patrick18287926

4 years

@arneelof @RolandDunbrack My professor. 100% warrior👌

0

1

Patrick Bryant

@Patrick18287926

2 months

@RolandDunbrack Ligand format. Can't be included in protein chains sometimes and sometimes not. Sometimes multiple ligands are merged, sometimes they are too similar to amino acids, sometimes the residue number doesn't change for the atoms, sometimes it does.

0

1

Patrick Bryant

@Patrick18287926

2 months

@FurmanLab I don't think it matters if the structures were included in training/not if you still use them at inference. Would be surprised if AF3 doesn't manage given the native structures as input

0

1

Patrick Bryant

@Patrick18287926

9 months

@MethembeThomas Yes, a MSc is needed to apply.

0

1

Patrick Bryant

@Patrick18287926

11 days

@BensenHsu Great summary 👍

0

1

Patrick Bryant

@Patrick18287926

9 days

@BioBoiTW Thanks 🙏

0

1

Patrick Bryant

@Patrick18287926

2 years

@arneelof Update: we have now added the possibility to upload your own receptor structures and to specify a starting sequence to optimise further. This also makes it possible to only predict by setting NITER=1. If you have a request, contact us. 😊

0

1