Researchers at Harvard Medical School have created an artificial intelligence (AI), model that can find treatments which can reverse disease in cells. This is a big breakthrough that could change how new drugs are discovered.
The new model, called PDGrapher and available for free, looks at many factors that cause disease and finds the genes most likely to help sick cells become healthy again. Unlike usual methods that check one protein or drug at a time, PDGrapher considers more possibilities to find better treatments.
The program also finds the best single or team of targets for therapies that can reverse disease. This study, which got some government funding, was published in Nature Biomedical Engineering on Tuesday.
By focusing on the best targets to reverse disease, this new way could speed up finding and making new drugs, helping with conditions that have been hard to treat using traditional methods, according to the researchers.
”Traditional drug discovery is like trying many different dishes to find one that tastes perfect,” said Marinka Zitnik, the study’s lead author and an associate professor at the Blavatnik Institute at HMS.
”PDGrapher works like a master chef who knows exactly what the final dish should be and how to mix the right ingredients to get that taste.”
How PDGrapher works: Mapping complex connections and effects
PDGrapher uses a type of artificial intelligence called a graph neural network. This tool not only looks at individual pieces of data but also studies the connections between these pieces and how they affect each other.
Also see: International Day to Protect Education From Attack: A Call for Safe Schools in Nigeria
In biology and drug discovery, this method is used to map the connections between different genes, proteins, and signaling pathways in cells. It predicts the best mix of treatments that can fix a cell’s problems and return it to a healthy state.
Benefits of the new model
The researchers made the tool using data from diseased cells both before and after treatment. This helped the model figure out which genes to target to change cells from sick to healthy.
They then tested it on 19 datasets covering 11 different types of cancer. They used both gene-based and drug-based methods. The tool was asked to predict treatment options for cell samples it hadn’t seen before and for cancer types it hadn’t been trained on.
The tool correctly found drug targets that were already known to be effective. These targets were deliberately left out during training to make sure the model wasn’t just memorizing the right answers. It also found more targets that have recent evidence supporting their use.
