Thomas Guiziou
@cen0ten.bsky.social
PhD Student in the Vorobieva Lab in de novo protein design
on peut combattre fermement l'antisémitisme d'ED et être intraitable sur de l'antisémitisme à gauche. Les deux peuvent exister en même temps, même, l'un ne concerne pas vraiment l'autre
December 2, 2025 at 9:44 PM
on peut combattre fermement l'antisémitisme d'ED et être intraitable sur de l'antisémitisme à gauche. Les deux peuvent exister en même temps, même, l'un ne concerne pas vraiment l'autre
If one side is right, then in ~10 years they will be rewarded by exactly that, and they cannot rush things.
2/2
2/2
December 2, 2025 at 1:28 PM
If one side is right, then in ~10 years they will be rewarded by exactly that, and they cannot rush things.
2/2
2/2
My feeling about this is that the contentious field becomes clearer with time. History will be the judge of all, and thus fighting over a "scientific win" on a short timescale is not that relevant.
1/2
1/2
December 2, 2025 at 1:28 PM
My feeling about this is that the contentious field becomes clearer with time. History will be the judge of all, and thus fighting over a "scientific win" on a short timescale is not that relevant.
1/2
1/2
While there is a lot of work and proof that the model produce good designs, to me it is unclear wether that method outperform previous methods.
December 2, 2025 at 11:05 AM
While there is a lot of work and proof that the model produce good designs, to me it is unclear wether that method outperform previous methods.
About the tests, we have optimization of known binder, failed binder from RFdiffusion and design of binder in known well-behaved fold (NTF2, TIM-barrel). There is also unconditional protein binder design, but to well known target. 3-
December 2, 2025 at 11:05 AM
About the tests, we have optimization of known binder, failed binder from RFdiffusion and design of binder in known well-behaved fold (NTF2, TIM-barrel). There is also unconditional protein binder design, but to well known target. 3-
One puzzling idea of the HalluDesign paper is the Fig 1b. Authors claim that smoothing the sequence-structure landscape with their approach helps to produce better deisgns. Yet that is an unrealistic representation of the underlying space that the model is suppose to explore. 2-
December 2, 2025 at 11:04 AM
One puzzling idea of the HalluDesign paper is the Fig 1b. Authors claim that smoothing the sequence-structure landscape with their approach helps to produce better deisgns. Yet that is an unrealistic representation of the underlying space that the model is suppose to explore. 2-
Is it a preprint or an advertisement?
Because the design methodology seems quite thin
Because the design methodology seems quite thin
December 2, 2025 at 10:24 AM
Is it a preprint or an advertisement?
Because the design methodology seems quite thin
Because the design methodology seems quite thin
Always impressive to see what the Fleishman Lab achieve with Rosetta !!
2/2
2/2
November 30, 2025 at 10:09 PM
Always impressive to see what the Fleishman Lab achieve with Rosetta !!
2/2
2/2
what for ? If I code as well as the llm, why need to use it
November 29, 2025 at 11:07 AM
what for ? If I code as well as the llm, why need to use it
By only using LLM one risk to only be as good as the tool. Which to me feels like a liability on the job market
November 27, 2025 at 10:02 PM
By only using LLM one risk to only be as good as the tool. Which to me feels like a liability on the job market
the question is more, what do you learn when you use the LLMs ?
Sure you boost your short term productivity, but what about long term understanding and productivity ?
Personally I need to understand very well the basic to be able to build on top of it.
Sure you boost your short term productivity, but what about long term understanding and productivity ?
Personally I need to understand very well the basic to be able to build on top of it.
November 27, 2025 at 10:01 PM
the question is more, what do you learn when you use the LLMs ?
Sure you boost your short term productivity, but what about long term understanding and productivity ?
Personally I need to understand very well the basic to be able to build on top of it.
Sure you boost your short term productivity, but what about long term understanding and productivity ?
Personally I need to understand very well the basic to be able to build on top of it.
In the pursuit of the origin of life, the definition of life is not important. But in the pursuit of a strong theoretical base for biology, the definition of life should be important. Which I think should concern most of biologist.
November 26, 2025 at 3:37 PM
In the pursuit of the origin of life, the definition of life is not important. But in the pursuit of a strong theoretical base for biology, the definition of life should be important. Which I think should concern most of biologist.
Yet, for biologist, it should be important to define what we are working on. One would assume it is essential to have a good theoretical base to the science.
November 26, 2025 at 3:26 PM
Yet, for biologist, it should be important to define what we are working on. One would assume it is essential to have a good theoretical base to the science.
Cristals fit in this definition and are not considered alive by most, so I would agree
November 26, 2025 at 3:08 PM
Cristals fit in this definition and are not considered alive by most, so I would agree
very nice figure, is it from the chroma paper ?
November 16, 2025 at 9:36 PM
very nice figure, is it from the chroma paper ?
yes, and the same can be said about artists
November 16, 2025 at 8:45 PM
yes, and the same can be said about artists
one of my former PI did his primer design on word document and kept all of his plasmid sequences on word documents
November 16, 2025 at 4:20 PM
one of my former PI did his primer design on word document and kept all of his plasmid sequences on word documents
It should be it isn't, thus we would hold said tool to a higher standard than our own research which does not make sense.
When publishing a paper, do we estimate the carbon footprint or pollution ? And then compare it to the societal gains ? When we do we can then ask the same to other.
When publishing a paper, do we estimate the carbon footprint or pollution ? And then compare it to the societal gains ? When we do we can then ask the same to other.
November 7, 2025 at 12:16 PM
It should be it isn't, thus we would hold said tool to a higher standard than our own research which does not make sense.
When publishing a paper, do we estimate the carbon footprint or pollution ? And then compare it to the societal gains ? When we do we can then ask the same to other.
When publishing a paper, do we estimate the carbon footprint or pollution ? And then compare it to the societal gains ? When we do we can then ask the same to other.
while making the experiments behind the paper you don't have a carbon footprint ?
November 7, 2025 at 12:13 PM
while making the experiments behind the paper you don't have a carbon footprint ?
Do you provide the carbon footprint behind your paper ?
I am not a fan of qed but this is not a valid issue.
I am not a fan of qed but this is not a valid issue.
November 7, 2025 at 11:34 AM
Do you provide the carbon footprint behind your paper ?
I am not a fan of qed but this is not a valid issue.
I am not a fan of qed but this is not a valid issue.
maintenant c'est tout droit pour la 6ème étoile
October 31, 2025 at 11:57 AM
maintenant c'est tout droit pour la 6ème étoile
It would make our life a bit easier if it was the case.
But I think the data derives from the principles and not the other way around.
We have to have a hypothesis or a theory to analyze the data. Otherwise we do unsupervised DL and we don't learn much, from my pov at least.
But I think the data derives from the principles and not the other way around.
We have to have a hypothesis or a theory to analyze the data. Otherwise we do unsupervised DL and we don't learn much, from my pov at least.
October 16, 2025 at 10:04 AM
It would make our life a bit easier if it was the case.
But I think the data derives from the principles and not the other way around.
We have to have a hypothesis or a theory to analyze the data. Otherwise we do unsupervised DL and we don't learn much, from my pov at least.
But I think the data derives from the principles and not the other way around.
We have to have a hypothesis or a theory to analyze the data. Otherwise we do unsupervised DL and we don't learn much, from my pov at least.