Designing accessible immunotherapies with AI

“Detecting unhealthy cells is one of the main jobs of the immune system, but it doesn’t always notice subtle signs of cancer or viral infection,” says David Baker. Though his lab has just published work in Science which he hopes can change that.  

The Baker lab used its generative AI protein design tools to create novel binders for peptide-MHC complexes to overcome these natural limitations of the immune system. As while T-cell receptors (TCRs) are designed to recognise foreign antigens presented on the MHC, the human body doesn’t have enough TCRs to recognise all the different antigens that could potentially be presented. And even if a compatible TCR is available, it might not bind with high enough affinity or be present at the right time and in the right place to actually enable killing of cancerous or infected cells.  

The recent Nobel laurate emphasises, “Our study shows that computer-designed proteins can help human immune cells pinpoint the right targets and work more effectively, potentially improving the precision and power of immunotherapies."

The work was spearheaded in the Baker lab by co-first authors Bingxu Liu, Nathan Greenwood and Julia Bonzanini, who were all inspired by the transformative potential of immunotherapy, but frustrated by issues with accessibility, given limitations due to necessary personalisation and associated cost.

These high-affinity, high-specificity binders have the potential to be used for both protein and cell-based therapies. As these binders are small, they are relatively stable and easy to produce, and subsequently transport. High-specificity is key to avoiding off-target effects, given that neo-antigens may differ only very slightly from those presented by normal cells. The fully computational design approach leverages RFdiffusion and ProteinMPNN, both readily available on the Baker lab website.

Julia is a third year PhD student in the bioengineering program at the University of Washington. After her undergraduate degree in biomedical engineering at Dartmouth College, she worked at an antibody engineering company. Julia reflects, “We were making these really cool antibody therapies, a lot of them for cancer. But I had family members back in Brazil who had cancer, and none of those big therapies that are best sellers in the US were even available for them.”

Julia emphasises her motivation to pursue a PhD in the Baker lab, “I was interested in finding alternative ways of providing care and making therapies that could be a lot more affordable and accessible on a global scale. I think de novo binder design has a lot more to offer there.”

Bingxu is originally from a village in central China, “Growing up there, and then later moving to a city, you really see how technology can change people's lives. That deeply motivated me to spread technology to people who are underrepresented and to make a therapy that can really benefit the whole population.”

Bingxu did his undergrad in China before moving to the US to get his PhD in immunology at MIT where he was co-advised by Nir Hacohen and Darrell J Irvine. He reflects, “I was actually a medical school dropout − I realised there's a lot more to do on the basic research side.” He continues, “My PhD allowed me to understand what the immune system is good at, and what is still needed to empower it to overcome the repertoire limitation.”

In collaboration with the lab of fellow team MATCHMAKERS member Chris Garcia, they solved the crystal structure of one of their binders together with the peptide-MHC complex. This allowed the team to demonstrate the accuracy of their computational pipeline, with near identical results from the computational and experimental structural approaches. They could then go on to design binders against antigens with no known experimental structures, and within mere days.

"Because our approach is fully computational, it can be applied to a wide range of disease markers, including those associated with currently untreatable conditions,” explains Nathan. And going completely computational saves both time and money, highlighting a potential path to scaling personalised immunotherapies.  

Bingxu reflects on the contributions of MATCHMAKERS, “The deep collaboration of different skill sets from MATCHMAKERS team members has been critical. The team’s antigen discovery is so beneficial for us, our teammates are building an incredible antigen database for us to keep working on.”

In the paper the lab focused on antigens presented by class I MHCs and showed the utility of their approach against a range of targets. But class I alleles are highly polymorphic, meaning patients would still need testing to first identify their HLA type, in order to find the right binder therapy for them. However, the lab aspires to design against non-polymorphic alleles in the future.

Julia emphasises the potential impact, “If you develop a therapy targeting those, it will be applicable for the whole population.”

The Baker lab and team MATCHMAKERS are on a mission to address the T-cell receptor challenge and make immunotherapy approaches both more successful and more accessible.

The team is excited to see how the community responds. As Bingxu puts it, “We’re showing the technology is ready to develop therapies, and we’re publishing the paper now to tell people this is possible. Now, let's discuss how to make it into a reality.”

Written by Rebecca Eccles with thanks to Bingxu Liu, Julia Bonzanini, Nathan Greenwood and David Baker

Read the paper: Design of high- specificity binders for peptide–MHC- I complexes

This work was published back-to-back in Science with two other papers, with similar approaches, also utilising the Baker lab AI tools:

De novo-designed pMHC binders facilitate T cell–mediated cytotoxicity toward cancer cells

De novo design and structure of a peptide–centric TCR mimic binding module*

*This work led by Chris Garcia was also delivered as part of team MATCHMAKERS

Read the perspective on all three papers:

Beyond the native repertoire

The Baker and Garcia lab papers were delivered as part of Cancer Grand Challenges team MATCHMAKERS, funded by Cancer Research UK, the National Cancer Institute in the US and The Mark Foundation for Cancer Research.

The Baker lab paper was also funded by The Audacious Project at the Institute for Protein Design, Howard Hughes Medical Institute, Parker Institutefor Cancer Immunotherapy, Washington Research Foundation, Children’s Hospital of Philadelphia, Microsoft, the US Department of Defense (HDTRA), National Science Foundation (2045054), and National Institutes of Health (2R01AI103867-11, P30 CA008748, R35 GM142795, R35 CA241894, R50 CA265328).