Even before Generative AI caught fire, computationally aided drug discovery was an area of intense interest for researchers. It should be: among other things, AI is a powerful tool for sorting through many possible solutions to a problem, and biology is well known for having enormous solution spaces requiring thorough analysis. Even after the hard work of identifying a dysfunctional protein or pathway whose modification might heal a patient, the number of molecules or proteins that could be used for the purpose is mind-boggling. A 2019 study published in Nature suggests humans may be capable of producing as many as one quintillion (1018) unique antibodies.1
We found in Antiverse a strong team using a combination of computational and biological innovations to look through that universe of antibodies efficiently, and with the audacity to apply their technology to challenging GPCR targets with potential impact on significant diseases. At Innospark, we get excited about companies using AI for good. Antiverse undoubtedly is.
The problem of drugging GPCRs
Let’s back up to the problem again. Many drugs work by getting in the way. By taking precisely the right shape, they bind to a problematic bodily protein to inhibit its action, stopping a chain of events. Form equals function , and finding exactly the right molecular shape is the challenge. This is true both for small molecules, chemicals synthesized in a lab and often delivered in little white pills, as well as for protein-based medicines, the comparatively large molecular structures often manufactured with the help of engineered cells and delivered by injection.
Antiverse works on antibodies, a type of protein-based medicine, so we’ll focus on those. To inhibit a bodily protein target, an antibody must bind to it strongly. The strength of this binding affinity makes the antibody potent even in small quantities. The antibody must also bind only to the intended target, lest it also go off and interfere with other bodily functions it isn’t intended to. In other words, antibodies must also have high specificity in order to avoid off-target effects.
Achieving both high affinity and high specificity – along with making sure an antibody has other nice drug properties – is difficult enough when designing an antibody to bind a protein target that’s naturally free-floating, but biotechnology has developed tools to help. The target protein can be produced in concentrated solutions, allowing for a screening process that gives lots of potential antibody binders a chance to attach and identify themselves as strong candidates. However, the challenge becomes much harder when some of biotechnology’s screening tools can’t be used, which is exactly the case with GPCRs.
GPCR stands for G Protein-Coupled Receptor. GPCRs are a class of membrane proteins that transmit signals, and your body has more than 800 known examples, related to a number of diseases. The GPCR sits embedded in the cell membrane, poking out either side, and its key job is to transmit messages into the cell. This often looks like the GPCR binding a signaling molecule outside the cell, and then changing shape in a way that conveys the message inside the cell. The message is received, but the cell’s protective membrane remains intact. Neat trick. But because GPCRs sit in the cell membrane - like icebergs in the sea, only partially exposed - they’re much more challenging to find antibody binders for. We can take the GPCR protein out of its cell membrane and put it in a concentrated solution, but if we do that the GPCR falls apart and looks nothing like its natural shape. Antibodies that attach to the exposed and deformed GPCRs in this solution will likely fail to recognize the GPCR’s native shape back in the context of the body. So, if we’re going to find antibodies to bind to GPCR targets, we need to 1) choose antibodies that bind to the small portion of the GPCR that’s naturally exposed above the cell membrane, and 2) test those antibodies against GPCRs in their natural membrane-bound position.
The Antiverse solution
We’re excited about Antiverse because we believe they have answers to both problems. Co-founders Murat Tunaboylu and Ben Holland lead an impressively multidisciplinary team bringing together AI with structural and molecular biology expertise in their solution.
Antiverse solves the problem of binding to a small piece of a GPCR with an interesting AI implementation. Rather than trying to bind to the GPCR target using a random set (or ‘library’) of antibodies, Antiverse’s machine learning models design a set of antibodies specific to that target from scratch. These models, trained from years of in-house experience, create a library of antibodies much more likely to contain a strong binder for the target than a random library might.
To solve the second problem of testing prospective binders against GPCRs in their natural place in the cell membrane, Antiverse brings to bear a biological innovation. Ordinarily, GPCRs are present on the surface of cells in relatively low numbers – numbers so low that even high affinity antibodies may pass by cells with GPCRs on the surface and not stick, simply for never having bumped into one. However, Antiverse has developed proprietary cell lines that can be designed to populate their surface with millions of copies of a target GPCR, allowing prospective antibody binders to be tested properly whilst maintaining the natural structure of the GPCRs.
With the combination of their computational and biological solutions, we hope and expect Antiverse will be able to design strong antibody binders to GPCR targets much more quickly and successfully than has previously been possible. In fact, by designing antibodies against specific and challenging sections (or ‘epitopes’) of already challenging GPCRs, one industry expert shared that Antiverse was doing something that previously could not be done.
Into the AI-enabled Future of Drug Design
Successfully designing antibodies able to bind to GPCRs and other difficult targets gives Antiverse the means to make many new medicines. Those more than 800 GPCRs are implicated in diseases across oncology, immunology, and more, and many still need better therapies. We know current approaches are inadequate based on the number of GPCR programs pharma companies have shelved or abandoned, and we’re encouraged by the interest the industry has shown in Antiverse’s technology. Murat, Ben, and their diverse team of biologists and ML scientists have the expertise to bring an AI + biology solution to this historically challenging problem. They embody Innospark’s belief in AI used for good, and we’re excited to support them.
References:
Commonality despite exceptional diversity in the baseline human antibody repertoire. Nature. https://www.nature.com/articles/s41586-019-0879-y