Last week, we announced that Recursion, in collaboration with malaria expert Dr. Elizabeth Winzeler at UC San Diego, is working on a new project: using artificial intelligence to find new treatments for malaria.
We’re thrilled to embark on this endeavor to tackle one of the world’s most enduring medical challenges. The project represents an opportunity for the Recursion team to advance our capabilities in drug discovery in important ways that will inform our work moving forward. First, malaria is an infectious disease, caused by the Plasmodium parasite. The advances we make as we seek treatments for Plasmodium infection will empower us to more broadly discover new medicines for other infectious diseases. Second, because malaria affects over 200 million individuals worldwide, the lessons that we learn, particularly lessons pertaining to translating our discoveries to patients, will enable our team to find medicines for other diseases that affect much larger populations.
Despite enormous efforts by researchers around the world, malaria is a particularly difficult problem to untangle. Although some kinds of malaria can be effectively treated, certain strains of malaria have the ability to persist in the human liver, preventing the disease from being eradicated.
Malaria is a worldwide problem. The map above shows areas at highest risk of malaria transmission in red.
Recursion’s Approach to Malaria
Liver-stage malaria can be treated with drugs targeting either the parasite or the host. Very few relevant targets are known, however. Recursion’s approach is ideally suited to finding treatments in situations like this.
We don’t need to know specific molecular targets; instead, we can test a library of drugs on malaria-infected human liver cells and apply artificial intelligence approaches to images of those cells to figure out which drugs restore infected cells to a state of health. This means that the drugs we discover have the potential to more broadly act on the cell, because they reverse all of the cellular features of disease, rather than acting on a single target or pathway.
Malaria is spread by mosquito bite, posing challenges to the prevention of malaria transmission.
Host Biology’s Role in Malaria Infections
Also breaking from tradition, we’re not primarily focused on affecting the Plasmodium parasite. As with any infectious disease, the host is indispensable for the survival and propagation of liver-stage malaria, which means that the host’s biology could potentially be manipulated to treat or prevent the infection.
This is important because the many strains of malaria are so diverse, and like many infectious agents, malaria has the potential to rapidly evolve resistance to antimalarial drugs targeted at its own biology. Host-factors are therefore compelling targets in infectious disease. Since we are looking at the phenotypes of infected human cells, our platform is ideally suited for identifying treatments that assist the host in responding to infection, rather than directly targeting the bug.
Charting New Biology in Infectious Disease
Another exciting aspect of Recursion’s malaria project is that it’s adding to our vast database of relatable biological image data. To develop any drug, it’s important to know the mechanisms, both helpful and harmful, that a particular drug is engaging in a human host, but this is largely uncharted space.
Because we’ve collected such a massive dataset, encompassing the responses of billions of cells to thousands of drugs and hundreds of diseases, we can build models to predict the mechanisms of any compound we screen, regardless of whether or not it has been previously well-studied.
Recursion’s approach to drug discovery involves a unique combination of automated experimental biology and artificial intelligence applied to cellular images.
Not only will we find drugs for malaria, but this approach also has the potential to apply to lots of other infectious agents. For example, we could potentially identify therapeutics that have the ability to extend the utility of antibiotics through empowering host cells to fight bacterial infection, boosting the body rather than killing the bug.
Target Agnostic Approach to Malaria Drug Development
Our target-agnostic approach and ability to identify drugs that empower host factors are already pretty exciting, but this project is actually even cooler than that. No matter how well-laid-out a project is, biology is biology, and there are bound to be surprises. As we’ve encountered some of these surprises, we’ve determined some pretty interesting things. For example, we’ve found that we can identify cells that have been infected with the malaria parasite without looking at the parasite itself.
When we initially started the malaria project, we expanded our Cell Painting protocol to include an extra stain so that we could visualize the parasite — this also meant extending the spectrum of visible light that we typically use, which required considerable effort and the use of a Plasmodium antibody that’s in short supply. We then had to adjust our machine learning algorithms to segment out the parasite, and extract not only the ~1,000 population-level morphological features we typically profile in human cells, but also features associated with the parasite. Still more challenging, the infection rate of malaria in our human liver cell cultures is pretty low (less than 1% of cells become infected), so our data scientists didn’t have much to work with.
Single-Cell Profiling To Dissect Host Response
To overcome this, our data scientists worked hard to adapt our algorithms and brought single-cell profiling online. This enabled us to use the parasite as a landmark to identify infected cells, and then profile just that subpopulation. We could then compare the infected subpopulation to the uninfected subpopulation without looking at parasite features — and we found that those two populations are distinct. Not only is this an interesting finding that we’re following up on to decode the biology of cellular host responses to malaria, but we also now have the ability to examine how drugs affect the phenotypes of individual infected cells relative to their uninfected neighbors, rather than having only a population-level view.
Malaria-infected human hepatoma (HepG2) cells that have been stained and imaged by Recursion. Plasmodium schizonts appear in yellow.
A Mission to Improve Lives
Recursion’s first foray into infectious disease is most exciting to me, however, because of the potential positive impact on patients. Although the cause of malaria is very clear, it’s an underserved indication with an urgent unmet medical need — and we are experts at working in underserved indications (e.g. our rare genetic disease work). Identifying treatments with the potential to eradicate liver-stage malaria is great proof-of-concept, not only because it’s an application that urgently needs quick- and inexpensively-discovered drugs, but also because it’s requiring us to build out the toolkit that we’ll need to tackle other infectious diseases.
It’s my hope that, ten years down the road, we’ll have identified not only several new treatments for malaria that are actively used in the field, but we’ll also have addressed a huge swath of viral, bacterial, and other infectious disease indications, both large and small, that lead to an estimated 15 million deaths every year — and in so doing, we’ll have radically improved the lives of patients suffering from these diseases.