Fighting antimicrobial resistance with AI and synthetic biology

An AI-based pipeline for de novo engineering of novel bioactive peptides to address the global health emergency of antimicrobial resistance

Yoshua Bengio - Co-Lead, Scientific Director of Mila, professor at Université de Montréal

Mike Tyers - Co-Lead, Principal Investigator at IRIC, professor at Université de Montréal

Antimicrobial resistance (AMR) is an on-going, building pandemic that encompasses myriad microbial pathogens including bacteria, fungi, viruses and parasites. Resistance to all known antimicrobials is documented, mutating variants AMR are now rampant in the developing world and threatening us all. AMR pathogens currently cause an estimated 700,000 deaths per year, with many more deaths caused by infectious pathogens that have not yet acquired resistance. The O’Neill report[1; see supplemental material for references] predicts this will rise to 10 million deaths per year by 2050 if AMR continues to spread unchecked, with estimated economic losses of 100 trillion USD by 2050. The entire global healthcare infrastructure faces collapse due to AMR, from fatal acute and chronic infectious disease to lethal secondary complications of cancer therapy and surgical intervention. Despite this dire situation, the pharmaceutical industry has largely abandoned antimicrobial drug development because new drugs are only used for patients with no other options, yielding an unviable revenue model. Philanthropic and government investment in continuous antibiotic discovery is required to solve the AMR problem. Our solution will ensure that the antimicrobial drugs and AI methods developed are openly accessible to all countries, particularly developing nations. 

Given the widespread prevalence of AMR, and the many diseases complicated by antibiotic resistance, our solution targets the global community, especially low and middle-income countries plagued by bacterial diseases that are increasingly resistant to antibiotics. In alignment with the One Health concept emphasizing health interrelationships between humans, animals and the environment, our solution may also be deployed to improve animal health and reduce zoonotic sources of AMR. AI-based rapid and systematic development of AMPs against major existing and emerging pathogenic species in humans and livestock will help enable the containment of AMR and its eventual elimination as a threat against global health. We will focus initially on development of AMPs targeting major Gram-negative enterobacterial pathogens since these diseases affect large populations in developing countries. We will also target Staphylococcus aureus, the main Gram-positive AMR pathogen, and a rapidly emerging fungal pathogen Candida auris, both of which are prevalent in hospital settings. AMR in these pathogens is spreading rapidly and a major long-term threat to humanity. Once AMPs are validated in preclinical animal models we will work with governmental and non-governmental agencies, including the Gates Foundation, GARDP and DNDi, in order to translate our work to human populations.

Our solution will provide five primary contributions to public good. First, we will provide open source code for our machine learning algorithms so that other researchers can build on our approach and re-use our concepts and/or code in other health applications. Second, we will publish all of our work in an unrestricted open-access format including all of the experimental datasets used to predict and optimize novel AMPs. Third, we will also provide all information, including unpublished datasets, as a web-based resource for AMP discovery. Our datasets will also be submitted to open access data repositories to maximize dissemination. Fourth, we will make our engineered probiotics available to interested parties including other academic research groups, not-for-profit and commercial partners interested in livestock applications, and commercial partners interested in moving our AMPs into human clinical trials. Fifth and finally, the project will train students and postdoctoral fellows to become experts in this important interdisciplinary field, who will themselves contribute to AI and drug development for public good in the future. We will ensure that principles of equity, diversity and inclusion are applied in all recruitment decisions and in the workplace environment.

Our solution will create tangible deliverables and impacts. Most centrally, we will deliver an open-access AMP discovery platform applicable to many different microbial pathogens, particularly infectious diseases that plague developing nations with massive health, societal and economic impacts. Diseases caused by pathogenic E.coli, Shigella and iNTS kill millions worldwide each year, primarily young children in African nations. We believe that conventional drug discovery paradigms have failed in this context and that a revolutionary solution is needed to address these worsening problems. It is for this reason that we have chosen to focus our AI-driven platform on high priority enteropathic bacterial pathogens as a primary test case. As we develop AMPs active against enterobacteria and other pathogens, we will engage with the Bill and Melinda Gates Foundation and non-profits such as DNDi and GARDP to initiate early-stage clinical trials in humans and field tests in livestock. We have been funded by the Gates Foundation on other projects and are currently in discussion with the Foundation for other AI-driven applications that target infectious disease, including projects on COVID-19. We will also approach partners to position our platform as an alternative to conventional antibiotic use in animal husbandry.

Our AI-driven solution is inherently scalable to all microbial infectious diseases, which impact many millions of lives globally, especially in the developing world. As detailed in the O'Neill report[1], without a dramatically improved antibiotic discovery pipeline, the AMR crisis threatens to completely overwhelm global health in the coming decades, at a cost of many trillions of dollars and millions of lives lost each year. Our solution also aligns with the One Health concept by replacing conventional antibiotic use with AMPs attuned to livestock pathogens, thereby mitigating a primary current source of AMR. The first year will focus on the discovery of AMPs against multiple critical pathogens that kill hundreds of thousands each year. With optimized AMPs, the second year will focus on testing these AMPs in validated animal models, including implementation of novel synthetic biotic delivery methods, while continuing the discovery of new AMPs and AMP combinations against different pathogens. Beyond the two-year term, our platform will be scalable to other microbial diseases caused by bacteria, fungi, viruses or parasites. Our platform also has the potential to dramatically scale economic benefits through the use of engineered probiotics to dramatically cut manufacturing costs and delivery logistics.

We will first validate to what extent the predicted effectiveness of AMPs obtained using our current algorithms matches the experimental data we propose to acquire. Then we will validate our algorithms for learning to search in the space of AMPs by comparing them with simpler and more established methods for searching in such a space, e.g., random search, evolutionary search or with a standard supervised learning predictor trained on the assay data and guiding a local search. Finally, we will validate whether the combinations of AMPs proposed by the AI system are as effective as predicted.

For the experimental perspective, we will use established growth inhibition assays to quantify AMP activity and the iterative improvement of activity. A further important measure of impact will be the identification of many different AMPs with different sequences, demonstrating that we can rapidly find new AMPs needed to combat AMR. The differential activity of AMPs against different pathogens will show that we can find pathogen-specific AMPs that have minimal collateral effects on the host microbiome. Finally, the demonstration that AMPs can mitigate symptoms or prevent death in mouse infection models will position the platform for trials in livestock animals and humans.

We do not foresee technical barriers to achieving our goals because all AI and biology methods are in place. The expertise of the AI and biology teams will  ensure that any minor technical issues will be surmountable. Traditional concerns with peptide-based drugs will be mitigated by recent breakthroughs in peptide stabilization and delivery technologies, which we will incorporate as needed. Besides the need for funds to perform the required experimental evaluations and analyses of the proposed algorithms, an important challenge is the high level of multidisciplinarity, with synthetic biology and drug discovery expertise on one hand and machine learning expertise on the other hand. Fortunately, we have been working together on several projects in recent years and are able to communicate effectively to identify and resolve problems. However, this is a continuous challenge since newcomers in our respective teams have to be appropriately trained. To address this, we use extensive documentation, modern communication tools that facilitate quick information exchange and cross-pair team members from different research domains. Downstream of the project, the real-world trials will require engagement of key partners but we are well positioned through existing interactions with the Gates Foundation, DNDi, GARDP and collaborators in developing nations.

Université de Montréal

McGill University

Mila Quebec AI Institute 

IRIC (Institute for research in immunology and cancer) at U. Montréal

Polytechnique Montréal

Michael G. DeGroote Institute for Infectious Disease Research at McMaster University

There is an urgent need to develop a fast and continuously renewable pipeline for new antimicrobials in order to stave-off the global threat of AMR. Our preliminary evidence suggests that integrating the tools we have built in the AI and synthetic biology domains has outstanding potential to dramatically shift the AMR landscape. We therefore seek funding to demonstrate the full potential of AI-driven AMP discovery and make the technology freely available to as many others as possible. Given the high-stakes race against AMR, we believe that an urgent investment in such technology is required. The Trinity Challenge aims to address problems on the scale of the global AMR crisis, unlike most conventional funding agencies. Since drug development to combat AMR is unlikely to be tackled by the pharmaceutical sector due to understandable business model constraints, opportunities such as the Trinity Challenge are the best avenue to tackle the threat of AMR. The barriers we aim to overcome include the experimental demonstration of the proposed AI-driven approach for AMP discovery, proof in animal infection models that designed AMPs have the potential to mitigate AMR, and the translation of our findings to human populations who are most at risk.

We have the explicit strong support of our institutions (U. Montreal, McGill U., Mila, IRIC, McMaster) each of which is committed to fostering AI-driven drug discovery. Our institutions  provide us with space, infrastructure (including expensive computing resources needed here) and logistical support. We expect to continue partnerships with the Bill and Melinda Gates Foundation (BMGF) and the Weston Family Microbiome Initiative on strategies to tackle infectious disease and AMR. We note that BMGF is a founding member of the Trinity Challenge. We are also seeking support for related aspects of AI-driven drug discovery from the National Research Council of Canada and we are working with colleagues at DNDi and GARDP for downstream collaborations on clinical trials if our project is successful. We have previously been supported by Genome Canada and Genome Quebec for early work on our engineered probiotic platform, and we will explore funding possibilities for delivery of AMPs in human and animal health contexts. Finally, the Canadian Institute for Advanced Research (CIFAR) funds the AI research chairs for Bengio, Bacon Precup and Chandar that have enabled much of our recent AI research, and encourages us to explore important social and health implications of AI in this project.