AI approach identifies potential new drugs to treat COVID-19

There is an urgent need to identify effective treatments for COVID-19. Currently, the University of California (UC) team at Riverside is working with many other academic and industry researchers to identify new drug candidates. They have developed a machine learning drug discovery pipeline to identify hundreds of new potential drugs that could be useful in treating COVID-19.

A drug discovery pipeline is a type of computational strategy linked to artificial intelligence, a computer algorithm that learns to predict activity through trial and error and improves over time.

The work is published in the journal HelioIn a paper titled,Predict new drugs for SARS-CoV-2 using machine learning from over 10 million chemical spaces“

Efforts to recycle drugs such as rendecibir have met with some success. There is no guarantee, but a vaccine against the SARS-CoV-2 virus could be months ahead.

“As a result, Dr. Ananda Sankalley, professor at UC Riverside, said, “As a result, as a first step towards the systematic discovery of new drugs to treat COVID-19, pipeline drug candidates like the one we have developed are It is very important to pursue: “Existing FDA-approved drugs targeting one or more human proteins that are important for viral entry and replication are now a priority for diversion as new COVID-19 drugs. There is a high demand for additional drugs or small molecules that can interfere with both SARS-CoV-2 entry and replication in the body. We have developed a drug pipeline.”

Ray’s lab graduate student Joel Kowalewski uses a small number of known ligands for 65 human proteins known to interact with SARS-CoV-2 proteins, including the ACE2 receptor. did.

We then trained a machine learning model to predict inhibitory activity and used them to screen FDA-listed chemicals and approved drugs (about 100,000) and about 14 million available chemicals.

“These models are trained to simply identify new small molecule inhibitors and activators (ligands) from their 3D structures,” Kowalewski said.

Therefore, Kovalevsky and Ray were able to create a database of chemicals whose structures were predicted as interactors of 65 different protein targets. They also assessed the safety of chemicals.

“The 65 protein targets are highly diverse and are associated with many additional diseases, including cancer,” Kovalevsky said. “Aside from ongoing drug diversion efforts towards these goals, I was also interested in identifying new chemicals that are currently poorly studied.”

Ray and Kowalewski use a machine learning model to screen over 10 million commercially available small molecules from a database of 200 million chemicals and 65 human proteins that interact with the SARS-CoV-2 protein. Identified the best-in-class hit of

We went one step further and identified compounds from hits already approved by the FDA, such as drugs and compounds used in food. We also used a machine learning model to calculate toxicity, which allowed us to reject potentially toxic candidates. This helped prioritize chemicals predicted to interact with the SARS-CoV-2 target. Their method not only allows them to identify the highest scoring candidates with significant activity against a single human protein target, but also finds some chemicals predicted to inhibit more than one human protein target. I was able to.

“The compounds most excited about my pursuit are those that are predicted to be volatile and set up the unusual potential of inhalation therapy,” Ray said.

“Historically, the treatment of diseases has become more complex as the understanding of the disease deepens and how individual genetic diversity contributes to the progression and severity of symptoms,” Kovalevsky said. He said. “Machine learning approaches like ours can play a role in predicting the evolving therapeutic landscape by providing researchers with the potential for further research. This approach relies heavily on data. Yet, virtual screening helps researchers to ask new questions and discover new insights.”

Ray and Kowalewski argue that computational strategies for initially screening a large number of chemicals are superior to traditional cell culture-dependent assays, which can be expensive and can take years to test. I am.

“Our database serves as a resource for quickly identifying and testing new safe treatment strategies for COVID-19 and other diseases that are associated with the same 65 target proteins,” he said. “The pandemic of COVID-19 was what motivated us, but the predictions from over 10 million chemicals not only accelerate COVID-19 but also drug discovery in the fight against many other diseases. I am expecting.”