Deep learning targets drug combinations for COVID-19


Using computational biology to identify individual therapeutics that could prove effective in the fight against COVID-19 is already a daunting task with billions of candidate molecules. Throw drug combinations into the mix, and the compound compounds make the task exponentially. Researchers have used deep learning to shorten the lists and identify promising drug combinations faster from COVID-19.

“Unfortunately, there are two main challenges that prevent you from using existing deep learning approaches to predict therapeutic chemical combinations for emerging pathogens such as SARS-CoV-2,” the researchers say wrote. “First, deep neural networks require a large amount of training data with measured synergy values. While such data are readily available for some diseases such as cancer, the amount of SARS-CoV-2 drug combination data (5) is very limited. ”

Second, they wrote, “Even the largest combination screen for cancer only covers about 100 different molecules as the number of paired combinations grows quadratically.”

To overcome these obstacles, the researchers developed a deep learning architecture called ComboNet, which “models the molecular structure and biological targets together to predict synergistic drug combinations,” hypothesizing that explicit modeling of drug-target interactions reduce the need for extensive data on combination synergy.

“By modeling drug-biological target interactions, we can significantly reduce reliance on combination synergy data,” says Wengong Jin, postdoc at MIT’s Broad Institute and lead author of the study. “In contrast to previous approaches, which use the drug-target interaction as fixed descriptors, our method learns to predict the drug-target interaction from molecular structures. This is an advantage as a large part of the compounds contain incomplete information on drug-target interactions. ”

Researchers evaluated ComboNet in a number of ways, including testing in silico Reuse of 30 tested combinations. Through these tests, they identified two combinations with strong ones in vitro Synergy: remdesivir and reserpine as well as remdesivir and IQ-1S.

“In general, ComboNet represents an advance in predicting new chemical-chemical synergies for cases where minimal combination training data is available,” the authors write. Next, they hope to feed in additional data on protein interactions and gene regulatory networks. They also plan to explore the use of active learning, which will help reduce the biases of the models towards certain chemical spaces and increase accuracy.

The research discussed in this article was updated in the September 2021 issue of. published as “Deep Learning identifies synergistic combinations of active ingredients for the treatment of COVID-19” PNAS. The paper that is available here, was written by Wengong Jin, Jonathan M. Stokes, Richard T. Eastman, Zina Itkin, Alexey V. Zakharov, James J. Collins, Tommi S. Jaakkola, and Regina Barzilay.


Leave A Reply