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.
> 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.
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.