Clinical AntiMicrobial Resistance Forecasting System

I will integrate large public datasets to forecast the future clinical antimicrobial resistance landscape. Predictions are based on observations, of which combination of: geographic, climatic and genomic circumstances were associated with past mobilization and transmission of non-human AMR genes to human and clinical relevance.

Patrick Munk
Assistant Professor
National Food Institute
Technical University of Denmark

Infectious diseases account for roughly 14% of all human deaths and the global disease landscape is changing with anthropomorphic climate changes. Examples include ecosystems changing, which alters transmission patterns of diseases, and host ranges of vectors in vector-borne diseases. Increasing antimicrobial resistance (AMR) in pathogens were predicted to grow the annual AMR death toll from 700K in 2016 to 10M in 2050, only losing out to heart diseases. The most comprehensive analysis to date, estimated 1.27M deaths due to AMR in 2019 alone. While a global issue, AMR is predicted to hit LMICs more severely. In some rare instances, we have identified a likely environmental source of new clinical AMR genes, but for most we do not.

The project will forecast ARGs that do spill into human bacteria from animals and the environment, and which of those might reach clinical significance. Which geographies, livestock and wildlife hold the ARGs that will emerge as the future clinically relevant ARGs? Predictions of the future clinical AMR will help the pharmaceutical industry design the right drugs for the right bugs, can help focus the increasing global genetic surveillance efforts and enable the design of effective interventions in high-risk areas.

As AMR is a global problem and this project seeks to limit the flow of non-human ARGs into humans and clinics on a global scale, the stakeholders are many, but policy makers and pharmaceutical industry will be the most important.

Using the DTU National Food Institute status as WHO Collaborating Centre on AMR, I will report the findings and recommendations to the WHO, who can integrate it in their global priorities. I trust that public sharing of my results on cross-sectorial ARG mobilization hotspots and low-resolution areas, will help prioritize interventions and more targeted surveillance. 

The project output will be of primary interest also to the pharmaceutical industry and researchers on antibacterial therapy. Having estimates of the needed “drug-bug” combinations of tomorrow will allow them to prioritize which products to develop and where they will be needed. I will therefore contact and meet with stakeholders of pharmaceutical industries. Copenhagen where the PI stays, has major players in this field, and also has antimicrobial production. I will set up an online project LinkedIn group and use my network to invite global stakeholders to discuss and meet to gather feedback and shape the project.

The research findings produced by this project will be published as open access peer-reviewed journal publications. We will also host a project website with datasets and visualizations, which could include transmission risks choropleths, bar charts of risk groups, and transmission network graphs. Data presented here will be useful for non-technical personnel including policy makers.

Developed software including the predictive network model will be made freely available to non-profit organizations through a git repository. Datasets shared on the website will be useful for academic researchers who wish to build and extend on the project.

For-profit organizations like pharmaceutical industries will be allowed to purchase licenses to those datasets and the code to run modified models in private. If such licenses are sold, it will help fund the continued operation and updates of our servers, code base, the project website and the datasets.

Carbapenem resistance killed ~ 430,000 people globally and is most commonly caused by the blaNDM which has clinically circulated since 2008. One way to save lives would be through decreasing the frequency of human introduction events, which could clearly save a lot of lives by keeping resistance at bay for one more year. Theoretically, it should be more efficient to intervene at rare cross-sectoral events than trying to stop an ARG after it is already disseminating through e.g. human gut Proteobacteria and their composite plasmids.

If we can confirm the hypothesis from Global Sewage results that SubSaharan Africa is a mobilization hotspot and it has global reach, our results should help convince richer countries that new investments to close down transmission routes has a high return-on-investment.

Another potential avenue comes from preparing for an event. Knowing which ARGs have historically jumped to human relevance, means that the pharmaceutical industry and other researchers can focus on developing the “right drugs for the right bugs” to impact future health.

Scientific publications will be advertised widely. I will also offer talks on the subject on the annual “Bestil en forsker” (rent a scientists) where institutions can book me for talks at no cost. I would advertise this offer specifically to the pharmaceutical industry.

• Y1: Developing a project-specific communication plan while growing my networks and audience reach on SoMe

• Y1: Create a closed, online stakeholder group to engage with throughout the project for feedback

• Y2: Active communication campaigns about project e.g. before and during the AMR awareness week

• Y3: Making press releases and SoMe posts targeted politicians and pharmaceutical industry around each project publication and combined results

I plan for six peer-reviewed publications to come out of the project. This will be the primary measure that I plan to measure success against. Research publications will detail both the technical developments, code produced, datasets and biological implications.

 I will compare and monitor the research impact first and foremost via traditional measures like number of citations, but also from more varied sources like Altmetric on my studies (https://www.altmetric.com/details/139554172).

The number of downloads and citations to the processed AMR datasets deposited in Zenodo will also be measured.

Currently, I am experiencing large costs associated both with actual compute and data storage in our shared Danish supercomputer system (Computerome2), where projects are billed by resources used in each project. Costs of course reflect both the high level of professionalism, support employees, security (less needed for publicly available data), data backup (wasteful to pay for raw, publicly deposited data) and service agreements ensuring almost no downtime (less important for multi-year projects).

To overcome these cost constraints, I am working on acquiring funding for upgrading and adding to our self-managed genomic epidemiology server cluster in the DTU server rooms. Both the number of high-core count compute nodes, GPUs, and storage arrays need upgrades for us to continue to address large-scale genomic epidemiology questions in a cost-effective manner.

Most of my research is specifically on using large-scale genomics and data science to understand AMR globally. I have worked with collaborators in more than a hundred countries, including many LMICs where we helped provide the first public microbiome and resistome datasets.

I want to scale up our analyses and be able to work on integrating large public datasets to understand the global One Health epidemiology of antimicrobial resistance. That requires both more computational power, and salary for me and a few colleagues to devote our time.

I would like to work with the Antimicrobial Action Fund, Pfizer and GlaxoSmithKline to learn if and how research projects like this could fit into the ‘Netflix model’ ‘push’ incentive of subscription contracts requiring pharmaceutical industries to make needed antibiotics available for the market.

After the research project, and for long-term sustainability, we would like to work with the ENA on hosting a server node that runs our bioinformatic analyses and downstream dataset updates. Our group has previously worked with Guy Cochrane on some of our bioinformatic solutions. It would be a more secure long-term solution if we would run out of funds for operating our own server cluster sometime in the future. And having the computational tool live on the same local network as the stored data, means less inefficient download of all datasets over North Sea cables connecting UK and Denmark.