This is an awesome technology. There is vast good it can do in medicine, biotech, materials science, etc.
But how concerned should we be with novel proteins being pumped into the biosphere? What does each new design do to our immune systems and ecosystems? How robust are these systems to, in effect, any protein that can be printed, and any substance that they can catalyze?
I am skeptical that a regulatory regime could reliably predict the consequences of releasing novel proteins via simulations or small scale experiments.
“Printing proteins” is a routine procedure since the late 70s and it included novel proteins or variations of
existing proteins since the early days of recombinant synthesis. Insulin and growth hormones were some of the first practical applications.
Modern techniques mess with the print system of some cells to generate huge variability for specialized experiments. These proteins themselves are largely harmless or only viable in controlled conditions.
The type of research you can be a bit worried about is the one that involves designing entirely new contageous and airborne-transmissible viruses or perhaps other designer pathogens. If you only ever built a super hazardous protein, you would still need to find a virus to deliver it somewhere practical (or inject it, like insulin, if it works outside of cells) and then it falls to a different type of problem anyways. Viruses often only need to multiply to cause harm, so the dangerous parts of them are their injection, attachment, replication, or immune evation pieces and there is enough examples of these known from the study of existing viruses, so you dont really have much need of this type of computational tool.
Are we just ignoring the potential for totally novel forms of prion-like proteins? Like there are many other proteins. If another of them is prone to catalyzed misfoldings like prions are, that could be a seriously humanity-threatening event.
Not sure what you mean. One can easily just make the existing prions if they wanted; they are good enough to destroy humanity if one finds a potent delivery mechanism. Again, the problem is not the protein printing tech or the protein design tech; you need to figure out a contagious, multiplying delivery machanic, ie a virus.
It may surprize people who dont work in these areas, but finding ways to kill humans by injecting something in them is completely trivial. The hard part is finding something that does not kill them and has a therapeutic effect.
prions are transmitted from animal to animal in the wild, for example chronic wasting disease or kuru. Isn't it possible for them to spread via an STD or something? Or can they only be grown in the brain?
If that was a genuine question, yes I have a very decent idea what prions are. We were discussing if the computational design tech should be giving people pause and the answer is no. If anyone really wants to create nasty stuff they dont need computers much and certainly no fancy ML tools. After the design, getting nasty stuff into humans typically happens either via viruses, pathogens, foods, or illicit drugs. Illicit drugs and foods are monitored in many parts of the world, and in Europe it is hard to mess with the food supply at scale. But even if you did, there is really no need for these designer tools as there are enough nasty sequences, including prions, that are known. On the other hand, these computational tools can help create safe biomolecules — it is just too easy to create nasty stuff with chemistry and biochemistry, but it is still very very hard to create safe and effective therapeutic molecules.
> Viruses are unnecessary for prions (near as we can tell), and a nasty virus can do plenty of direct damage.
Agreed
> Someone could (if they discovered one) construct a nasty prion directly eh? And we’d probably be pretty fucked.
It would be easy to construct a potent prion-like sequence that works in isolation (in vitro) with such a tool, but not clear what the rationale would be even if one has nefarious purposes. We know of enough of them (eg pick Creutzfeldt–Jakob disease for human hosts; if you have a hydrogen bomb, the variations on the engineering could be intersting but most likely not needed.) I dont think we would be fucked if someone constructed new prion-like sequences with this method, though we might be at risk if someone makes genuine efforts to find ways to deliver the existing (or new ones) to humans at scale. The main risk factors in prions diseases involve exposure and host compatibility. You need enough time and thus enough of a starting dose for harm to happen. You need to enter the host and go to the right place, say nervous system, without being chopped up along the way by random proteases and without self aggregating in the wrong place and becoming useless. There may be other computational tools that in combination would help realise more potent/nasty/risky designs, but it is probably much easier to use human cell lines and benchwork (design a funky evolution experiment) to fish out nasty sequences that actually work in human cells.
Overall in biology and chemistry killing is easy, therapy is hard. Prions are somewhat interesting from a biology or engineering perspective, but they dont have the huge added risks of rapid exponential growth and easy transmission to hosts that many viruses have. They seem more like a curiosity and primitive danger form, and not a huge public health risk factor. Maybe there even exist some semi benign (even useful?) forms related to prions that we dont study much yet and they may end up influencing things like our understanding of aging or immune responses.
Prions relate to nucleation and first order phase transitions, so they have some technical appeal to people with past background in physics or chemistry, including myself, but they are not making it in my top 20 threats for humanity, not even if there were concerted efforts by rogue actors to create new nasty ones.
> and there is enough examples of these known from the study of existing viruses
So making any arbitrary protein literally go viral is a solved problem? Yes, I can see how that adds spice to a tool that supercharges the exploration of protein-space.
Yes it is technically a solved problem. But no you dont need to try to add a novel protein if you had nefarious intent. There exist more than enough well studied and fully understood examples.
Some modern food variants contain highly immunogenic proteins. Gluten protein complexes in modern wheat are much more likely to trigger autoimmunity compared to those from ancient wheat varieties. Same is said about certain synthetic enzymes in cold-water detergents and allergy, so this is a valid concern, but it predates ML-guided design of proteins.
This is an awesome technology. There is vast good it can do in medicine, biotech, materials science, etc.
But how concerned should we be with novel proteins being pumped into the biosphere? What does each new design do to our immune systems and ecosystems? How robust are these systems to, in effect, any protein that can be printed, and any substance that they can catalyze?
I am skeptical that a regulatory regime could reliably predict the consequences of releasing novel proteins via simulations or small scale experiments.
(Nearly) not concerned at all.
“Printing proteins” is a routine procedure since the late 70s and it included novel proteins or variations of existing proteins since the early days of recombinant synthesis. Insulin and growth hormones were some of the first practical applications.
Modern techniques mess with the print system of some cells to generate huge variability for specialized experiments. These proteins themselves are largely harmless or only viable in controlled conditions.
The type of research you can be a bit worried about is the one that involves designing entirely new contageous and airborne-transmissible viruses or perhaps other designer pathogens. If you only ever built a super hazardous protein, you would still need to find a virus to deliver it somewhere practical (or inject it, like insulin, if it works outside of cells) and then it falls to a different type of problem anyways. Viruses often only need to multiply to cause harm, so the dangerous parts of them are their injection, attachment, replication, or immune evation pieces and there is enough examples of these known from the study of existing viruses, so you dont really have much need of this type of computational tool.
Are we just ignoring the potential for totally novel forms of prion-like proteins? Like there are many other proteins. If another of them is prone to catalyzed misfoldings like prions are, that could be a seriously humanity-threatening event.
Not sure what you mean. One can easily just make the existing prions if they wanted; they are good enough to destroy humanity if one finds a potent delivery mechanism. Again, the problem is not the protein printing tech or the protein design tech; you need to figure out a contagious, multiplying delivery machanic, ie a virus.
It may surprize people who dont work in these areas, but finding ways to kill humans by injecting something in them is completely trivial. The hard part is finding something that does not kill them and has a therapeutic effect.
prions are transmitted from animal to animal in the wild, for example chronic wasting disease or kuru. Isn't it possible for them to spread via an STD or something? Or can they only be grown in the brain?
You seem to have no idea what prions are?
No virus is necessary for prions to spread, and they are nearly impossible to destroy. And merely ingesting the right ones is sufficient.
If that was a genuine question, yes I have a very decent idea what prions are. We were discussing if the computational design tech should be giving people pause and the answer is no. If anyone really wants to create nasty stuff they dont need computers much and certainly no fancy ML tools. After the design, getting nasty stuff into humans typically happens either via viruses, pathogens, foods, or illicit drugs. Illicit drugs and foods are monitored in many parts of the world, and in Europe it is hard to mess with the food supply at scale. But even if you did, there is really no need for these designer tools as there are enough nasty sequences, including prions, that are known. On the other hand, these computational tools can help create safe biomolecules — it is just too easy to create nasty stuff with chemistry and biochemistry, but it is still very very hard to create safe and effective therapeutic molecules.
Viruses are unnecessary for prions (near as we can tell), and a nasty virus can do plenty of direct damage.
Someone could (if they discovered one) construct a nasty prion directly eh? And we’d probably be pretty fucked.
So far though, we only know of a handful of actual working prions, so maybe there isn’t other others?
> Viruses are unnecessary for prions (near as we can tell), and a nasty virus can do plenty of direct damage.
Agreed
> Someone could (if they discovered one) construct a nasty prion directly eh? And we’d probably be pretty fucked.
It would be easy to construct a potent prion-like sequence that works in isolation (in vitro) with such a tool, but not clear what the rationale would be even if one has nefarious purposes. We know of enough of them (eg pick Creutzfeldt–Jakob disease for human hosts; if you have a hydrogen bomb, the variations on the engineering could be intersting but most likely not needed.) I dont think we would be fucked if someone constructed new prion-like sequences with this method, though we might be at risk if someone makes genuine efforts to find ways to deliver the existing (or new ones) to humans at scale. The main risk factors in prions diseases involve exposure and host compatibility. You need enough time and thus enough of a starting dose for harm to happen. You need to enter the host and go to the right place, say nervous system, without being chopped up along the way by random proteases and without self aggregating in the wrong place and becoming useless. There may be other computational tools that in combination would help realise more potent/nasty/risky designs, but it is probably much easier to use human cell lines and benchwork (design a funky evolution experiment) to fish out nasty sequences that actually work in human cells.
Overall in biology and chemistry killing is easy, therapy is hard. Prions are somewhat interesting from a biology or engineering perspective, but they dont have the huge added risks of rapid exponential growth and easy transmission to hosts that many viruses have. They seem more like a curiosity and primitive danger form, and not a huge public health risk factor. Maybe there even exist some semi benign (even useful?) forms related to prions that we dont study much yet and they may end up influencing things like our understanding of aging or immune responses.
Prions relate to nucleation and first order phase transitions, so they have some technical appeal to people with past background in physics or chemistry, including myself, but they are not making it in my top 20 threats for humanity, not even if there were concerted efforts by rogue actors to create new nasty ones.
> and there is enough examples of these known from the study of existing viruses
So making any arbitrary protein literally go viral is a solved problem? Yes, I can see how that adds spice to a tool that supercharges the exploration of protein-space.
Yes it is technically a solved problem. But no you dont need to try to add a novel protein if you had nefarious intent. There exist more than enough well studied and fully understood examples.
Some modern food variants contain highly immunogenic proteins. Gluten protein complexes in modern wheat are much more likely to trigger autoimmunity compared to those from ancient wheat varieties. Same is said about certain synthetic enzymes in cold-water detergents and allergy, so this is a valid concern, but it predates ML-guided design of proteins.