Drug development has changed dramatically in the past few decades. As technology and science continue working in tandem, bringing new life-saving drugs to market is faster than ever. But designing some drugs, specifically those which are antibody-based, remains quite challenging despite their incredible potential. Even so, roughly 30% of all biologics approved by the Food and Drug Administration (FDA) last year were antibody-based therapies.
That’s why the Vancouver, Washington, based biotech firm Absci Corporation is turning to generative artificial intelligence (AI) to improve the design workflow. Unsurprisingly, the same technology behind ChatGPT and other large language models excels at finding new antibodies that can treat disease with high accuracy.
The team at Absci is one of many exploring the technology’s potential for drug development. If its current progress is any indication, the future of antibody therapy research could be closer than anyone imagined.
Notably, Absci began as a drug manufacturing company in 2011. It has since refocused on drug discovery and is spearheading its operations with generative AI. The technology allows Absci to test nearly three million unique antibody designs every week—an unimaginable pace for antibody research just a few years ago.
Absci’s Chief AI Officer Joshua Meier said in a statement, “There’s a totally new paradigm for designing proteins. We are doing things we thought were impossible even five years ago. And today, a programmer can write code that 18-24 months later can save someone’s life.”
Antibodies, a unique type of protein that bind to specific receptors on other tissues or cells, are part of the body’s natural immune system. Unfortunately, they don’t always respond effectively to certain diseases or cancers. But once researchers identify the markers of a problematic cell or tissue, it’s possible to make new antibodies that bind to it and it alone. This opens the door for antibody therapies that treat or cure the condition by harnessing the power of the immune system.
Ensuring that antibodies are compatible with the body is challenging. Thanks to generative AI, though, Absci can quickly design antibodies with a high level of “naturalness,” or those which are likely to cause the desired immune response.
Perhaps more impressive, however, is the team’s ability to create antibodies that latch on to markers only when a cell shows certain traits—such as a mutation in cancer cells. This means the antibody latches onto only the affected cells and spares healthy ones. This is revolutionary compared to treatments like chemotherapy, which wreak havoc on diseased and healthy cells alike. Without AI, optimizing antibodies with such strict parameters takes years of testing for just one trial.
Absci CEO Sean McClain says, “You can’t go tell a mouse to develop an antibody that binds to a specific area of the target you want, that has the affinity that you want, the specificity and manufacturability that you want. But with our data, you are able to develop the right antibody the first time.”
Finding accurate, detailed data to train antibody-generating AI models has historically been a challenge for the industry. To enable its use of generative AI, the team trained its model on antibody data collected over the past decade. Thanks to its long-term commitment, Absci has been able to develop a “zero-shot” process to design antibodies for specific targets without using training data of similar antibodies. In other words, the model is free of bias and starts from scratch.
While this may seem counterintuitive, it actually leads to more diversity in the generated antibody models. By not relying on previous solutions, the algorithm is more creative and comes up with new approaches for the team to test. This is key to treatment breakthroughs in a world where humans have tried and failed to solve the same disease-related problems for years.
So, what comes next for Absci and its generative AI approach? The company envisions itself as the Google of antibody-based drug research. It is currently pursuing 17 research programs of its own as well as aiding other biomanufacturing and pharmaceutical companies in their research and design.
McClain has big hopes for the future. He says, “Fifteen years from now you would be able to take a patient sample, find the target that was relevant for that disease for the patient, and design an antibody to address that disease.”
“This is going to start to enable precision medicine in the future,” he adds.
Whether or not AI can bring us to that point in such a short timeframe remains to be seen. But the application of antibody-based drug research is a promising one for generative AI. As this technology emerges from its infancy and more developers learn to adapt its powerful tools to their needs, there is no limit to the possibilities that could arise.