Follow ZDNET: Add us as a preferred source on Google.
ZDNET’s key takeaways
- Harvard researchers designed a new AI model, PDGrapher.
- It can identify treatments to restore diseased cells to health.
- This could have larger impacts on drug discovery.
While AI’s most common use cases involve helping people with their everyday tasks, it can also go far beyond that, even helping make medical breakthroughs.
Also: Can AI outdiagnose doctors? Microsoft’s tool is 4 times better for complex cases
Last week, Harvard Medical School published a study on a new AI model called PDGrapher. According to Harvard, the model can analyze the connections between genes, proteins, and signaling pathways inside cells to identify the best combination of therapies that would effectively restore healthy cell behavior. This could enable new treatments for conditions that have previously been unable to be found via traditional methods.
The findings
In a summary of the study, which was partially federally funded, the authors explain that usual drug discovery approaches tackle one protein at a time, and work in cases such as kinase inhibitors — drugs that prevent cancer cells from expanding by blocking certain proteins — but can fall short when the disease involves interactions between multiple signaling pathways and genes.
“Traditional drug discovery resembles tasting hundreds of prepared dishes to find one that happens to taste perfect,” said study senior author Marinka Zitnik in the summary. “PDGrapher works like a master chef who understands what they want the dish to be and exactly how to combine ingredients to achieve the desired flavor.”
Also: AI could dull your doctor’s detection skills, study finds
The researchers trained the model on a dataset of diseased cells before and after treatment so that PDGrapher could use that data to identify which genes transform cells from a diseased to a healthy state. Then the model was put to the test on 19 datasets spanning 11 types of cancers, where it was asked to predict various treatment options without having seen the cell samples before.
The tool accurately predicted the drug targets known to work and identified other targets with clinical evidence to support them. PDGrapher also outperformed “other similar tools,” the authors wrote (though they did not specify any), ranking the correct therapeutic targets up to 35% higher and 25 times faster.
Also: AI’s not ‘reasoning’ at all – how this team debunked the industry hype
The researchers identified many ways PDGrapher can optimize drug discovery by identifying multiple targets that can reverse the disease. According to the post, this could speed up the process, streamline research efforts, and reduce instances in which complex diseases with multiple pathways, such as cancer, evade drugs. Currently, the team is using the model to tackle brain diseases, including Parkinson’s and Alzheimer’s.
AI in medicine
Though still new, AI has made several recent strides for medical use cases. Last year, for example, AI models’ tendency to hallucinate actually helped Stanford researchers find new drug compounds at an exponentially faster rate than they could have with basic computing alone. At the same time, however, studies show AI chatbot users are perhaps over-reliant on AI tools for medical advice, which can be factually incorrect or otherwise unreliable, and does not replace information from a medical professional.
PDGrapher is availabe via Github here.