When the Covid-19 pandemic struck in early 2020, doctors and researchers rushed to find effective treatments. There was little time for rest. “Making new drugs will last forever,” says Caroline Uhler, a computer biologist in MIT’s Department of Electrical Engineering and Computer Science and the Institute for Data, Systems and Society, and an associate member of the MIT Broad Institute and Harvard. “Really, the only appropriate option is to return drugs. ”
The Uhler team has now developed a machine learning-based approach to identifying drugs that are already on the market that may be reversible against Covid-19, especially the elderly . The system accounts for changes in gene expression in lung cells caused by both the disease and aging. Such a combination could allow medical experts to seek out drugs more quickly for a clinical trial in elderly patients, who tend to experience more severe symptoms. The researchers identified the RIPK1 protein as a promising target for Covid-19 drugs, and identified three approved drugs that work on the RIPK1 expression.
The research appears today in the journal Nature Communication. Co-authors include MIT PhD students Anastasiya Belyaeva, Adityanarayanan Radhakrishnan, Chandler Squires, and Karren Dai Yang, as well as PhD student Louis Cammarata from Harvard University and longtime collaborator GV Shivashankar from ETH Zurich in Switzerland.
Early in the pandemic, it became clear that Covid-19 harmed older patients than the younger ones, on average. Uhler’s team was wondering why. “The most common idea is the aging immune system,” she says. But Uhler and Shivashankar suggested an additional feature: “One of the main changes in the lung that occurs through aging is that it becomes harder. ”
The stiffening lung tension reveals different patterns of gene expression than in younger people, even in response to the same symptom. “Earlier work with the Shivashankar laboratory showed that if you stimulate cells on a harder substrate with cytokine, similar to what the virus does, they will actually turn on different genes,” says Uhler. “It simply came to our notice then. We need to look at aging with SARS-CoV-2 – what are the genes at the crossroads of these two pathways? ” To select licensed drugs that could work on these pathways, the team turned to big data and artificial intelligence.
The researchers entered the most promising transplant candidates in three broad steps. First, they generated a large list of drugs that could use a machine learning method called autoencoder. Next, they mapped the network of genes and proteins involved in both age and SARS-CoV-2 disease. Finally, they used statistical algorithms to understand a cause in that network, allowing them to identify “upstream” genes that caused blocking effects across the network. In principle, drugs that target these genes and proteins should be promising candidates for clinical trials.
To generate an initial list of drugs, the team’s autoencoder relied on two main datasets of gene expression patterns. One database showed how expression in different cell types treated a range of drugs already on the market, while the other showed how expression treated infection with SARS-CoV-2. The autoencoder manipulated the databases to highlight drugs that appeared to have an effect on gene expression as opposed to the effects of SARS-CoV-2. “This use of autoencoders was challenging and required basic insights into the operation of these cloud networks, which we developed in a paper recently published in PNAS,” notes Radhakrishnan.
Next, the researchers shortened the list of potential drugs by entering major genetic pathways. They mapped protein interactions involved in aging infectious pathways and Sars-CoV-2. They then identified areas of overlap between the two maps. That effort identified the exact gene expression network that a drug needed to target to fight Covid-19 in elderly patients.
“At this stage, we had an unknown network,” says Belyaeva, meaning that the researchers did not yet have to identify which genes and proteins were “upstream” (ie they have inhibitory effects on the expression of other genes) and which was “downstream” (ie their expression has changed with previous changes in the network). A suitable drug candidate would target the genes at the upper end of the network to reduce the effects of infection.
“We want to identify a drug that will affect these different genes downstream,” says Belyaeva. So the team used algorithms that find causality in interaction systems to turn their unknown network into a causal network. The final causative network identified RIPK1 as a target gene / protein for potential Covid-19 drugs, as it has several downstream effects. The researchers identified a list of permitted drugs that act on RIPK1 and may be able to treat Covid-19. These drugs have previously been approved for use in cancer. Other drugs also identified, including ribavirin and quinapril, are already in clinical trials for Covid-19.
Uhler plans to share the team’s findings with pharmaceutical companies. She confirms that a clinical trial can be performed before any of the drugs they have named are used for re-use in elderly Covid-19 patients, to determine efficacy. While this particular study focused on Covid-19, the researchers say their framework is extended. “I am very pleased that this platform can be applied generally for diseases or other ailments,” says Belyaeva. Radhakrishnan emphasizes the importance of gathering information on how various diseases affect gene expression. “The more data we have in this area, the better it can work,” he says.
This research was supported, in part, by the Office of Naval Research, the National Science Foundation, the Simons Foundation, IBM, and the MIT Jameel Clinic for Machine Learning and Health.