Machine Learning Approach to Finding Treatment Options for Covid-19 | MIT News

When the Covid-19 pandemic broke out in early 2020, doctors and researchers were in a hurry to find an effective treatment. I didn’t have much time. “Manufacturing new drugs will take forever,” said Caroline Uhler, a computational biologist at MIT’s School of Electrical Engineering and Computer Science and the Institute for Data Systems Society, and an associate member of MIT and Harvard’s Broad Institute. Stated. “Really, the only convenient option is to reuse an existing drug.”

Uhler’s team has now developed a machine learning-based approach to identify drugs already on the market that could be reused to fight Covid-19, especially in the elderly. This system explains the changes in gene expression in lung cells caused by both disease and aging. This combination allows healthcare professionals to more quickly find drugs for clinical trials in older patients who tend to experience more serious symptoms. Researchers have identified the protein RIPK1 as a promising target for Covid-19 drugs and identified three approved drugs that act on RIPK1 expression.

The study will be published in the journal today Nature Communications.. Co-authors include MIT PhD students Anastasiya Belyaeva, Adityanarayanan Radhakrishnan, Chandler Squires, Karren Dai Yang, Harvard PhD students Louis Cammarata, and long-term collaborators at ETH Zurich in Switzerland. Includes one GV Shivashankar.

Early in the pandemic, Covid-19 was found to be more harmful to older patients than younger patients on average. I wondered why Wooler’s team. “The general hypothesis is the aging of the immune system,” she says. However, Uhler and Shivashankar suggested additional factors. “One of the major changes in the lungs caused by aging is Harden.. “

Hardened lung tissue exhibits a different pattern of gene expression than young people, even when responding to the same signal. “Early studies by the Shivashankar lab showed that, like viruses, stimulating cells on harder substrates with cytokines actually turns on different genes,” Uhler said. .. “So that motivated this hypothesis. We need to look at aging with SARS-CoV-2 — what genes are at the intersection of these two pathways?” Acting on these pathways. To select potential approved drugs, the team turned to big data and artificial intelligence.

Researchers have focused on the most promising drug diversion candidates in three major steps. First, they used a machine learning technique called an autoencoder to generate a large list of possible drugs. They then mapped a network of genes and proteins involved in both aging and SARS-CoV-2 infection. Finally, they used statistical algorithms to understand the causality of the network, allowing them to identify the “upstream” genes that caused the cascading effect throughout the network. As a general rule, drugs that target these upstream genes and proteins should be promising candidates for clinical trials.

To generate an initial list of potential drugs, the team’s autoencoder relied on two major datasets of gene expression patterns. One dataset shows how expression in different cell types responds to different drugs already on the market, and the other dataset shows how to infect SARS-CoV-2. Showed if it reacts to. The autoencoder scrutinized the dataset, highlighting drugs whose effects on gene expression appear to counteract the effects of SARS-CoV-2. “This autoencoder application was challenging and required basic insights into the behavior of these neural networks. It was developed in a recently published treatise. PNAS“Radhakrishnan said.

The researchers then narrowed down the list of potential drugs by focusing on key genetic pathways. They mapped the interaction of proteins involved in aging and the Sars-CoV-2 transmission pathway. Next, we identified areas that overlap between the two maps. The effort has identified the exact gene expression network that the drug needs to target to combat Covid-19 in older patients.

“At this point we had an undirected network,” says Belyaeva. That is, researchers have not yet identified which genes and proteins are “upstream” (that is, they have a cascading effect on the expression of other genes) and which are “downstream” (downstream). That is, their expression is altered by previous changes in the network). The ideal drug candidate would target genes at the upstream end of the network to minimize the effects of infection.

“We want to identify drugs that affect all of these differentially expressed genes downstream,” says Belyaeva. Therefore, the team turned an undirected network into a causal network using an algorithm that infers the causality of interacting systems. The final causal network was identified as a potential Covid-19 drug target gene / protein because RIPK1 has numerous downstream effects. Researchers have identified a list of approved drugs that act on RIPK1 and may treat Covid-19. Previously, these drugs were approved for use in cancer. Other identified drugs, such as ribavirin and quinapril, are already in clinical trials with Covid-19.

Uhler will share the team’s findings with pharmaceutical companies. She emphasizes that clinical trials are needed to determine efficacy before any of the drugs they identify are approved for reuse in elderly Covid-19 patients. This particular study focused on Covid-19, but researchers say their framework is extensible. “I’m really excited that this platform can be applied more commonly to other infections and illnesses,” says Belyaeva. Radhakrishnan emphasizes the importance of gathering information on how various diseases affect gene expression. “The more data we have in this space, the better this can work,” he says.

This study was partially supported by the Navy Research Office, the National Science Foundation, the Simons Foundation, IBM, and the MIT Jameel Clinic for Machine Learning and Health.