NETWORK-based machine learning has been used to improve the accuracy of predicting patient response to immune checkpoint inhibitors (ICIs). ICIs have considerably improved the survival rate of patients with cancer, with fewer side effects and longer-lasting treatment benefits compared to traditional chemotherapy.
However, one major challenge remains, in that only approximately 30% of patients receiving immunotherapy experience benefits from its therapeutic effect, and there are a lack of effective tools to predict patient response to this treatment. Moreover, another significant challenge of precision medicine using immunotherapy is recognising markers from ICI-treated patients that can anticipate drug responses across multiple cancer patient cohorts.
A study led by Sanguk Kim, Department of Life Sciences, POSTECH, Pohang, Korea, analysed clinical results of more than 700 ICI-treated patients diagnosed with three separate cancer types: bladder cancer, melanoma, and gastric cancer, as well as the transcriptome data of the patients’ cancer tissues.
The team successfully discovered new network-based biomarkers, which were then used to develop artificial intelligence, which could predict the response to ICI treatment. This network-based method is built on observations that genes with similar phenotypic roles tend to colocalise in a specific region of a protein–protein interaction network. This tendency has been exploited to identify gene modules that are much more reliable in predicting phenotypic outcomes than using single gene-based approaches.
The researchers concluded that treatment response prediction utilising network-based biomarkers is more effective than that based on conventional ICI treatment biomarkers, such as ICI targets or tumour microenvironment-associated markers. This research provides valuable evidence that using a network-based machine learning approach to identify biomarkers can make solid predictions of the ICI treatment response in patients with cancer.
Looking at future developments in the field, the authors stated: “We envision that our work here opens up interesting new research opportunities for precision medicine using ICI treatment.” Such research opportunities could, for instance, include a semi-supervised learning approach to train machine learning models.
