The latest breakthrough comes thanks to researchers at the Scuola Normale Superiore in Pisa. Raimondi: “Transferring knowledge from the laboratory to clinical trials.”
What if Artificial Intelligence could identify the best drug for each individual cancer patient? Researchers at the Scuola Normale Superiore in Pisa have developed a model that promises to revolutionize personalized medicine, as their recent study published in Nature Communications shows. “This line of research,” explains Professor Francesco Raimondi, “is part of the THE program, the Pnrr-funded health ecosystem, with the goal of harnessing bioinformatics techniques and Artificial Intelligence algorithms to suggest new targeted therapeutic approaches, toward an increasingly personalized medicine. Not just finding the most suitable drug for a given disease, but more importantly for that individual patient.”
Translated with DeepL.com (free version)
> Here< the article published in “il Riformista.”
The team used extensive pharmacogenomics databases developed by international institutions such as the Wellcome Sanger Institute in Cambridge and the Broad Institute in Boston. These dataset include information on approximately oncology drugs and over 4,000 non-oncology drugs tested on hundreds of cancer cell lines. “With this data, we can train predictive AI algorithms to identify already approved or late-stage drugs that could be repositioned for new oncology indications,” Raimondi adds.
A crucial aspect of the research is the possibility of speeding up the process of identifying new therapies, reducing the time it takes to bring a drug from a discovery phase to the clinic. “We hope to significantly shorten the time it takes to validate and apply new cancer therapies” Raimondi says. Illustrating how the developed model work is Francesco Carli [ > HERE < the conversation at un caffè con THE], Ph.D. student in the Bioinformatics Laboratory: “Our model combines two approaches of Artificial Intelligence: predictive and generative. We use techniques of machine learning to analyze the ‘identity card’ of cancer cells, assessing which proteins are expressed and how these affect drug response. It’s a complex problem: we have to analyze about a thousand cell lines and over twenty thousand genes to predict drug sensitivity.”
But the real added value of the research lies in the interpretability of the model. “One of the biggest limitations of AI in biomedicine is the transparency of the algorithm’s decisions. Our model, on the other hand, is interpretable: we are able to identify key genes that influence the prediction of drug sensitivity, allowing us to formulate new biological hypotheses,” Carli points out. Generative artificial intelligence plays a key role in the process. “We use advanced language models to generate textual descriptions of the analyzed drugs and to identify the most relevant biological processes,” Carli explains. “This approach allows us to combine experimental data and biological knowledge, improving the accuracy of our predictions and facilitating the repositioning of existing drugs for new therapeutic indications.”
The importance of this research lies in its practical applicability: an approach that integrates bioinformatics, big data and Artificial Intelligence to provide personalized therapeutic solutions with a potentially revolutionary impact for precision medicine. Although the results are still being validated, the prospects are promising. “The ultimate goal,” Raimondi concludes, “is to transfer this knowledge from the laboratory to clinical trials, accelerating the path to discovering new treatments for cancer patients.
