NIAPU: network-informed adaptive positive-unlabelled learning for disease genes identification

Gene-disease associations are fundamental for the understanding of disease mechanisms and for the development of effective interventions and treatments. Identifying genes not yet associated with a disease due to lack of studies is a challenging task in which prioritization based on prior knowledge can result helpful. The computational search for new candidate disease genes may be eased by Positive-Unlabelled (PU) learning, the machine learning (ML) setting in which only a subset of instances are labelled as positive, while the rest of the data set is unlabelled. In this work, we propose a set of effective network-based features to be used in a novel Markov diffusion-based multi-class labelling strategy for putative disease gene discovery. The performances of the new labelling algorithm and the effectiveness of the proposed features have been tested on five different disease datasets using three ML algorithms. Such features have been compared against classical topological and functional/ontological features showing that they outperform the classical ones both in binary classification and in the multi-class labelling. Analogously, the predictive power of the integrated methodology in searching new disease genes has been found to be competitive against the state-of-the-art algorithms.