The New Era of AI-powered Protein Design | César Ramírez-Sarmiento  ## Metadata - Author: [[TED Talks Daily]] - Full Title: The New Era of AI-powered Protein Design | César Ramírez-Sarmiento - Category: #podcasts - URL: https://share.snipd.com/episode/2de140f7-5a3b-4906-af1e-4c6c66792036 ## Document Notes ## Highlights - **Proteins Are Programmable Machines** - Proteins are sequences of 20 amino acids that fold into 3D shapes to dictate function. - Designing sequences can change function and solve human problems faster than natural evolution. Transcript: César Ramírez-Sarmiento My name is Cesar Ramirez-Armiento. I'm based in Santiago, Chile. I'm a protein engineer and designer. Proteins are macromolecules which are composed of amino acids. They are made of 20 different types of amino acids. They are represented by letters. So you can imagine an alphabet of amino acids. And you can imagine that these amino acids are connected to each other like beads on a string. And so that allows for them to come together in different geometries. And so they get a shape. They get a three-dimensional structure that allows for them to dictate their functions. We have many different proteins with many different shapes that actually perform different biological functions in cells. They allow us to digest food. They allow us to transport ions for electrical signals to go through neurons. They allow for the expression of different genes that regulate how our cells or how our body responds. Proteins are the workforce of cells. They are like a toolbox for cells to do whatever they have to do. Proteins have been evolving for millions of years for performing functions that are important for cellular life. They have been perfected by nature to do what they do now. But when it comes to problems that are important for humankind, like plastic contamination, carbon dioxide, problems in health, we want to make them better. ([Time 0:03:28](https://share.snipd.com/snip/be80313e-ab32-43fb-8cd2-77c85b5445a5)) - **AI Boosts Protein Design Success** - AI has raised protein-design success rates from ~1% to roughly 10–20%. - That boosts the fraction of computational designs that show desired activity in the lab dramatically. Transcript: César Ramírez-Sarmiento In the last five years, we have seen breakthroughs in artificial intelligence for designing proteins that we never imagined. Allow us for designing new protein structures, new protein shapes that encode bespoke functions for solving all types of problems. Before the advent of AI, the success rate for protein design was about 1% or less, which means if you created 100 proteins with 100 different sequences, maybe one of them would work. Now with the advent of AI, we see about 10 to 20 percent. So if you now take your 100 sequences that you generated in the computer, about 20 of them will actually have the desired activity, and some of them will be actually better than the input Sequences of the protein of interest that you're working with. ([Time 0:05:26](https://share.snipd.com/snip/a82b6b5f-1904-480d-8a6c-41478b209dd9)) - **Enzymes Enable Circular Plastic Recycling** - AI-designed enzymes can decompose plastics into small molecules used to remake plastic. - This approach aims for circular recycling where plastic becomes feedstock for new materials. Transcript: César Ramírez-Sarmiento And so there's a few companies that are working now on degrading plastic. And these proteins that perform chemical reactions are called enzymes. And enzyme design is a problem of its own that is very difficult to tackle. But there's one company in France and there's another company in China that are working on developing with the use of AI different enzymes that can actually degrade different types Of plastics. And the idea behind it is that we are then using that plastic as a feedstock for creating new plastic afterwards. ([Time 0:13:00](https://share.snipd.com/snip/2d57d601-c819-4313-af5e-6213839ce019)) - **Turn Computational Designs Into Tested Proteins** - Design proteins in silico, convert designs to DNA, synthesize genes, express proteins, then test experimentally. - Use iterative cycles from computation to lab validation for real-world deployment. Transcript: César Ramírez-Sarmiento What you do in the computer is that you go from protein back to DNA. So you have different scenarios. And for them, you have different ways of working with them. And so the idea is that you can train artificial intelligence models on this information about the sequence of a protein so the sequence of amino acids that compose a protein or you can Train models on the structures of proteins or you can train them on both and then in the computer after like a few days or weeks of work you will have a set of different sequences that will Encode these structures or these different printing functions. And then what you do after that is that you backtrack those designs from printing information into DNA information, so another form of alphabet. And then you purchase those genes from these companies that synthesize your genes and then put them into different bacterial or animal cells for expression of these proteins and then Testing them out in the lab. ([Time 0:14:03](https://share.snipd.com/snip/17006467-c634-4cff-9995-c4b717b91728)) - **Snails Sparked A Research Path** - As a child César noticed snails eating plastic plant pots and later learned enzymes that degrade plastics exist. - That memory guided his research interest into plastic-degrading enzymes as a professor. Transcript: César Ramírez-Sarmiento Thing that happened when I was younger. I think maybe I was like eight years old or something like that. My mom, she had some like plastic pottery for plants that she was like growing in the garden. And she said that the snails were eating through them. I was like, there's no way that that can happen. But the idea remained there for a long time. I was thinking, oh, maybe it's possible. But I just remembered that memory, right? And then when I was in university, I was actually learning about biochemistry, like learning about proteins and enzymes and all of these things. And then we had a class about enzymes and how they perform like different chemical reactions. And one friend of mine said, oh, what if they degrade plastics? And I was like, well, that doesn't exist. But little did I know that enzymes that could degrade plastics were being discovered. So the first one was, I think, 2003 or something like that. So that turned into my research topic when I became a professor. ([Time 0:15:40](https://share.snipd.com/snip/760e9755-8de9-46bc-811e-f31dd538720d)) - **Use Enzymes For Sustainable Waste Treatment** - Replace harsh chemical processes by enzyme-based alternatives to treat crustacean waste from fishing. - Use biological solutions to convert industry waste into fertilizers and useful products sustainably. Transcript: César Ramírez-Sarmiento So we have a huge fishing industry, right? And that fishing industry for some of the crustaceans that we're like extracting from the sea, a lot of it's not consumable. A lot of it's waste. And then almost all of the technologies for reutilizing those waste from like crustaceans are typically treated with like very harsh chemicals. ([Time 0:17:05](https://share.snipd.com/snip/390ee9bd-01b2-4c55-ae64-867e808cf2b3)) - **Tiny Folded Proteins As Drugs** - AI Proteins is making tiny folded proteins that target specific cell proteins for therapies. - These miniproteins could become precise pharmaceuticals for allergies and cancer. Transcript: César Ramírez-Sarmiento To change or make better, do you think? I'm very excited about Chris Balls, another TED fellow, his company, AI Proteins. And what they're developing in his company is that this very, very tiny protein, very few amino acids, right? When you make them too short, they don't fold into like a shape. Imagine like if you have like a piece of rock and you want to fold it. If it's too short, you cannot fold it, right? But if you have like a piece that is long enough, then you can fold it. So proteins also fold upon themselves. And so they're making this very, very tiny proteins that actually fold upon themselves. So they have a shape and that shape is complementary to different target cells that have proteins that are involved in cellular processes that are related to different diseases that Can go from like different allergic reactions to even like treatments for cancer. So the idea is to develop like this very, very tiny proteins that you can use as a pharmaceutical, as a drug. ([Time 0:18:10](https://share.snipd.com/snip/849a03b3-9a7a-4b71-9958-3e8d61bd233f)) - **Designing Enzymes From New Structures** - New methods now allow designing entirely new enzyme structures, not just tweaking sequences. - That opens possibilities for inventing novel biochemical reactions from scratch. Transcript: César Ramírez-Sarmiento So I think that I'm seeing a lot of advances right now in making new enzymes. So enzymes are very difficult to make. They have very specific sites on the protein surface. So if you imagine that you have like a sphere as a protein, then they have like a little hole on the surface that is called the active site. And the chemical reaction actually happens in that site and that site alone, which means that there has to be very specific amino acids from your protein in very specific positions. So the methods that we had like a year ago were like pretty bad at the enzyme design. And so before we will take an enzyme that we know it has some activity for some thing that we want it to do. Like there's a lot of enzymes that do plastic degradation, so we will improve the sequence, but we will never change the structure. But nowadays there's a lot of methods that actually can afford that. And so the idea is that you can create new structures that have never seen before in nature of enzymes, of new enzymes, and that allows for thinking about new chemical reactions that We can create from scratch. So that's very exciting. ([Time 0:19:50](https://share.snipd.com/snip/ec178529-ca88-4507-bd69-b29d4984ff55)) - **Build Risk Assessment Into Biodesign Models** - Evaluate and disclose risks when releasing AI biodesign models and consider 'unlearning' harmful potentials. - Pair model release with expert oversight and safety evaluations by AI safety institutes. Transcript: César Ramírez-Sarmiento So a bunch of scientists, including myself and other very well-known scientists in the realm of artificial intelligence for everything about biology, not only about product design, Signed some guidelines that were called Responsible AI for Biodesign that indicate that we will do significant efforts to identify risk in the different models that we develop for Different types of biodesign using artificial intelligence. And then try to indicate those risks whenever we release the models or try to do what people call unlearning, which is try to make models to somehow not capture this harmful potential When you release them to the public. For now, you still need an expert scientist because they are not very easy to use. But if you combine them with these large language models that allow for having a conversation with your computer without having the expertise for creating something. ([Time 0:22:55](https://share.snipd.com/snip/ddc6a3b9-55f3-4348-94f2-3596e948bce6)) - **Global Opportunity For Leadership** - Countries beyond the US can lead AI-for-biodesign policy and funding, and Latin America is stepping up. - Chile and other nations are investing in regulation and capacity to take leadership roles. Transcript: César Ramírez-Sarmiento In? I mean, I think, yeah, there's an opportunity for other countries to lead it. There's efforts, I know, in Europe. Denmark is putting a lot of funding into AI for biodesign. The UK is also investing a lot of funding into that. There's efforts, at least in Latin America, to also, like, step on these things and try to take the lead. So my country, actually, Chile, like leading for a while an initiative to ascertain the risk and the ethical usage of artificial intelligence for different purposes. And so there is an opportunity for using artificial intelligence for protein design methods in the country and be like a leading country in Latin America for that. But yeah, since the U.S. Has been experiencing some changes in the last year, yeah, there's a lot of countries that have been like stepping up and trying to take the leading input in the South. ([Time 0:25:10](https://share.snipd.com/snip/cd412efe-c2c4-47ff-9501-9b725f58e520)) - **Fear Of Pandemics, Hope In Biotech For Climate** - César fears future pandemics and human forgetfulness but finds hope in biotech investment for climate solutions. - He expects many biotech startups to tackle greenhouse gases and plastic degradation soon. Transcript: César Ramírez-Sarmiento Don't know. I think that after the pandemic, I'm always scared about, oh, what's gonna come next? Right? So like, is there gonna be another pandemic like that? And how are we gonna to respond to that? Instead of like being concerned about like our capacity to respond to things, I'm more concerned about like human behavior, which is like, we're very forgetful. And we might forget that we were like, at least in our case in Chile, we were in quarantine for a while. And I don't want to be in that situation again. So what I'm hopeful for is that there's a lot of investment and interest also from scientists to work on solutions, biotechnological solutions for combating climate change. And plastic degradation is just one example, but there's a lot of efforts for trying to eliminate greenhouse gases. We know that there are some protein-based solutions and also some cell-based solutions, and there's a lot of interest for investing on them. So I'm very hopeful that in the future, and actually in the near future, we will see a lot of startup biotechs that are working on climate change and successfully. ([Time 0:28:24](https://share.snipd.com/snip/099d6fbe-cbc8-4691-b363-070729d24137))