In the realm of Machine Learning, inference over datasets with high-dimensional feature (input) vectors can be challenging. To address this issue, a new framework has been developed that utilizes synergistic interactions to extract significant features from a given dataset. More specific by extracting the most important features the goal is to build up a regression graph-based model to assess the severity of coronary stenosis in patients undergoing computed tomography coronary angiography (CTCA) via the prediction of the gold standard fractional flow reserve (FFR). This approach is novel as it is the first time synergy has been used to identify relevant features in a Gaussian Process (GP)-based regression task. Numerical tests on a real medical dataset, namely SMARTool, showcase the impressive merits of the novel method that judiciously utilizes synergistic interactions, over state-of-the-art feature selection techniques.