The existing threat of Covid-19 has highlighted the urgent need to develop practical treatments for new health problems. One of the luxury benefits of deep learning is the ability to change the landscape as long as it responds to viral threats and has access to the right data.
As with all new medical disorders, data often takes time to catch up, viruses take less time to slow down, and can quickly mutate to become resistant to existing drugs. Because of their nature, they pose difficult challenges. As a result, scientists at MIT’s Institute for Computer Science and Artificial Intelligence (CSAIL) and Jameel Clinic for Machine Learning in Health asked: How can we identify synergistic drug combinations suitable for the rapidly expanding SARS-CoV-2?
Data scientists typically use deep learning to choose drug combinations with existing large datasets such as cancer and cardiovascular disease, but of course, new diseases with limited data. Cannot be used for.
Without the necessary facts and numbers, the team needed a new approach. It’s a neural network with two hats. Because drug synergies are often caused by inhibition of biological targets (such as proteins and nucleic acids), models are new to jointly learn drug-target interactions and drug-drug synergies. Mining the combination. Drug target predictors model the interaction between a drug and a set of known biological targets associated with the disease of choice. Target disease-related predictors learn to understand the antiviral activity of drugs. This means determining virus yield in infected tissue culture. Together, they can predict the synergistic effect of the two drugs.
Using this approach, a combination of two new drugs was found. Remdesivir (currently approved by the FDA for the treatment of Covid-19) and reserpine, and Remdesivir and IQ-1S, which have been shown to be potent against the virus by biological assays.Research has been done Release In Minutes of the National Academy of Sciences.
“By modeling the interaction between the drug and the biological target, we can significantly reduce our reliance on combination synergistic data,” said Broad, a recent CSAIL PhD from MIT and Harvard. Wengong Jin SM ’18, an institute postdoc, said. He is the lead author of a new treatise on research. “In contrast to previous approaches that used drug-target interactions as fixed descriptors, our method learns to predict drug-target interactions from their molecular structure, most of the compounds. This is advantageous because it has incomplete drug-target interaction information. ”
The use of multiple drugs to maximize efficacy while reducing side effects is virtually ubiquitous in the aforementioned cancers and cardiovascular diseases, including other hosts such as tuberculosis, leprosy and malaria. .. Special drug cocktails can, very importantly, reduce the serious threat of resistance and sometimes the public threat (think of it as methicillin resistance). Staphylococcus aureus Many drug resistance mutations are known as “MRSA” because they are mutually exclusive. It is much more difficult for the virus to develop two mutations at the same time and then become resistant to the two drugs in combination therapy.
Importantly, this model is not limited to just one SARS-CoV-2 strain, but could be used for increasingly contagious delta mutants and other variants of concern. Only additional drug combination synergy data for related mutations are required to extend the effectiveness of the model for these strains. In addition, the team applied an approach to HIV and pancreatic cancer.
The team plans to incorporate additional information such as protein-protein interactions and gene regulatory networks to further refine biological modeling.
Another direction for future work they are looking for is called “active learning.” Many drug combination models are highly uncertain in their predictions because they are biased towards a particular chemical space due to their limited size. Active learning guides the data acquisition process and helps improve accuracy in larger chemical spaces.
Jin wrote a treatise with Jonathan M. Stokes, a banding fellow at the Broad Institute at MIT and Harvard. Richard T. Eastman, a scientist at the National Center for Advancing Translational Sciences. Gina Itokin, a scientist at the National Institutes of Health. Alexey V. Zakharo, Informatics Leader at the National Center for Advancing Translational Sciences (NCATS). James J. Collins, Professor of Biological Engineering at MIT. Tomi S. Jaccola and Regina Bargeley, professors of electrical engineering and computer science at MIT.
This project is supported by Abdul Latif Jameel Clinic for machine learning in health. Defense Threat Reduction Agency; Patrick J. Adams Foundation; DARPA Accelerated Molecular Discovery Program. Partly due to on-campus / off-campus research programs at the National Center for Advancing Translational Sciences within the National Institutes of Health.
Deep learning helps predict new drug combinations to fight Covid-19
https://scienceblog.com/525595/deep-learning-helps-predict-new-drug-combinations-to-fight-covid-19/ Deep learning helps predict new drug combinations to fight Covid-19