Vilya Inc.

The lab of Nobel laureate David Baker keeps pushing the limits of what’s possible in making proteins with computers, opening up new possibilities in using AI to make drugs. The latest advance arrived Thursday in Science , with a team of 21 scientists using AI models to make enzymes from scratch that perform some fairly complicated chemical feats. The technology isn’t yet as simple as pushing a button and getting a protein, but the progress shows these models rapidly improving in tackling tougher protein design problems. Though these AI-made enzymes aren’t quite as good as the best ones found in nature, it’s still a step ahead for the AI protein field. Running what is likely the hottest lab in the protein design world, Baker has been studying and designing proteins with software since the 1990s. But AI advances have driven an explosion in progress these past few years. That culminated in winning the Nobel Prize in Chemistry in October, paired with a growing list of startups and spinouts from his lab. Baker has co-founded over 20 companies, including the billion-dollar debut of Xaira Therapeutics last year, while his lab has been able to make minibinders, antibodies and now enzymes from scratch. In an interview with Endpoints News , Baker laid out possibilities for using these latest enzyme advances in drug discovery, especially to design site-specific enzymes that cut DNA or RNA strands. That could give scientists a new level of precision and control in designing gene-editing tools. “The bigger picture is we can now use deep learning, ML diffusion methods, to make really active enzymes,” Baker said, adding his lab is now working on making nucleases, or enzymes that cut nucleic acids, and thinking about base editors, another type of gene editing. This latest Science project was led by a trio of scientists in Baker’s lab — Anna Lauko, Sam Pellock and Kiera Sumida —who wanted to make new hydrolases, or a type of enzyme that acts like scissors in cutting molecules into two pieces. These are well-studied, common enzymes that have been found across nature, from eating up plastics to allowing our muscles to contract, relax and move. Hydrolases are seen as a hard challenge for the protein design field. Their activity requires getting a multistep process just right, which can be harder than getting a protein to glom onto a target. The team started out with little progress in searching for active enzymes by making and screening thousands of designs. That changed when another Baker lab group developed RFdiffusion, a diffusion model that creates proteins. “There was a long period where we didn’t make anything that worked,” Lauko said in an interview. “And then RFdiffusion basically became available to us because people in our lab were developing it. Almost immediately, we made far more progress than we had in the past three years by completely changing how we were doing the design.” Once they got access to RFdiffusion in the winter of 2022, the team changed their approach. Rather than starting with the scaffold of an enzyme and trying to mutate it to get what they wanted, they could basically tell the model what they desired. The AI model then created the possible proteins. Their paper details rounds of testing and iterating on better enzyme designs. That includes implementing another, new AI model called PLACER, which acted like a quality control check in vetting the AI-made proteins. PLACER judged how compatible they were for the desired chemical reaction, helping filter out ones unlikely to succeed and not worth making and testing on the lab bench. These AI-made enzymes aren’t as active as the best hydrolases found in nature, according to their lab results. At least for now, the AI models still aren’t coming up with better solutions than a billion years of evolutionary pressures. “They’re still not as good, and that’s another future direction: is trying to understand what’s missing there to make these better and faster,” Lauko said. Looking ahead, Pellock described two major areas they are focusing future research on: making plastic-degrading enzymes and exploring the therapeutic potential for AI-made enzymes with other functions. In the therapeutic world, Pellock said proteases, or enzymes that break down proteins or peptides, are an area to watch. A long-term hope is being able to degrade any protein in the human body by building the right enzyme, he said, which could lead to new, more precise therapies that could also tackle currently unreachable targets. A provisional patent to the PLACER model was licensed to the Baker-founded biotech startup Vilya, according to the paper. Vilya has licensed other AI models from Baker’s lab, including an exclusive license to RFpeptides , a diffusion model for macrocyclic peptide design.

As interest grows in the ability to leverage certain ringed molecules as drugs, Orbis Medicines has raised €90 million in new funding less than a year since the company unveiled its seed round. Orbis’ goal is to develop oral drugs for biologic targets using what are called macrocycles, which are rings made of the same building blocks as proteins but that are far smaller than typical proteins. The biotech’s focus is blockbuster targets that already have antibody drugs on the market, particularly in chronic diseases where patients might prefer to take a pill regularly rather than getting their treatment via injections or IV. The €90 million Series A was led by New Enterprise Associates. It included other investors such as Eli Lilly, Cormorant Capital, and the Export and Investment Fund of Denmark — a government-backed fund that backs Danish companies — as well as founding investors Novo Holdings and Forbion. While Orbis is still several years away from human studies, macrocycles have garnered renewed interest over the prospect of making small molecule drugs or pills, which are simpler to make and administer, that can go after highly specific targets usually only accessible using antibodies. Newly-minted CEO Morten Døssing said that the company went out after launching publicly to increase its seed round “a little bit.” The biotech announced a €26 million seed round from Forbion and Novo Holdings in February. “There was a significant amount of interest, and then that spurred this discussion, ‘Well, if we are raising, why don’t we just raise for something more substantial?’” said Døssing, who served as executive chair of Orbis’ board for the past three years as a partner at Novo Holdings, the investment arm of Novo Nordisk’s parent foundation. Macrocycles have long been the basis of drugs, including antibiotics and antifungals, but most were based on existing macrocycles found in nature. Startups developing these next-generation macrocyclic drugs are now leveraging newer technologies including high-throughput screening and machine learning to engineer and design different peptide rings. Orbis’ science comes from Sevan Habeshian and Bicycle Therapeutics co-founder Christian Heinis out of the Swiss Federal Institute of Technology in Lausanne. It’s also working with another biotech called Vivtex to screen the oral bioavailability of drugs. Orbis is using unnatural amino acids, drawing from a library of some 400 fully synthesized building blocks as opposed to the 20 protein building blocks found in nature. “The chemical diversity is much greater, but in a much smaller footprint,” said Døssing, noting that macrocycles using natural amino acids are typically eight to 12 amino acids in size, meaning they are large and not ideal for oral drugs. He added that Orbis is looking at targets primarily in immunology at this point, but noted that there’s interest in developing oral peptide drugs for cardiometabolic conditions — a category that includes obesity. He pointed to AbbVie’s recent $200 million acquisition of Nimble Therapeutics , a biotech that is likewise developing oral peptide drugs in immunology. Other macrocycle startups include Vilya — its name comes from “The Lord of the Rings,” and it was launched by Nobel Laureate and protein design pioneer David Baker and ARCH Venture Partners — and Circle Pharma, which last year raised a $54 million Series D . Merck last year also signed a small macrocycle deal with Unnatural Products. Orbis plans to use part of its Series A to build out a team in Copenhagen, where the company that also has a team in Switzerland announced a new site in October. Døssing said that Orbis currently has 15 employees but hopes to have around 40 workers by the end of the year.

Dive Brief:Orbis Medicines, a European biotechnology startup, has raised 90 million euros, or about $93 million, in a Series A round announced Monday that will fund development of oral peptide drugs.Orbis is focused on macrocycles, a class of medicines that blends the strengths of small molecule and biologic drugs. The company aims to construct oral macrocycles that work against disease targets already validated by successful injectable medicines, but hasnt specified which it will go after.Incubated by Novo Holdings, Orbis emerged from stealth last year with 26 million euros in seed funding. The company was co-founded by Sevan Habeshian and Christian Heinis, the latter of whom helped start Bicycle Therapeutics, a publicly traded macrocycle drug developer.Dive Insight:Biologic medicines are a huge and still-growing segment of pharmaceutical research. A report last year from Future Market Insights noted the market for biologics was expected to reach about $521 billion in 2024 and could more than double by 2034.Yet biologics are generally delivered by injections or infusions, making them less convenient than small molecule drugs that are taken orally. That limitation leaves an opportunity for drugmakers able to develop medicines that pack the power of biologics into a pill. Switching people from injections to oral drugs could also help with patient adherence, said Morten Graugaard Dssing, Orbis CEO and a partner at Novo Holdings.Morten Graugaard Dssing is a partner at Novo Holdings specializing in seed investments, and now CEO of Orbis Medicines.Permission granted by Orbis MedicinesOne solution that has gained traction in recent years are macrocyclic peptides, which are a newer twist on macrocycle drugs, dozens of which have won approval. Shaped like rings, macrocyclic peptides can slip into cells the way small molecules do, but are big enough to interfere with protein interactions, making them possible solutions for some undruggable targets. They also appear resistant to degradation by enzymes or stomach acids.Those characteristics may mean macrocyclic peptides retain biologic-like potency while administered orally, a goal Orbis and other companies like Vilya and Circle Pharma hope to achieve.Pharmaceutical companies have shown interest, too. Merck & Co. has in late-stage testing an oral macrocyclic peptide drug for heart disease it claims could rival marketed injectable medicines.There's a need for better technologies around making truly orally available macrocycles for a lot of these hard-to-drug targets and transitioning from biologics into an all-oral format, Graugaard Dssing said.Orbis work is based on research from Heinis and Habeshian that showed how to produce a library of macrocyclic compounds and engineer them so theyre smaller, and better absorbed by the body than existing drugs.The company hasnt disclosed which diseases or targets it is pursuing. Graugaard Dssing said Orbis platform could be used for any target and is right for any indication, broadly.Orbis' Series A round was led by NEA. Lilly Ventures, Cormorant Asset Management, Novo Holdings, Forbion and the Export and Investment Fund of Denmark were also involved. '