Prediction of cancer driver genes through network-based moment propagation of mutation scores

Abstract

Motivation Gaining a comprehensive understanding of the genetics underlying cancer development and progression is a central goal of biomedical research. Its accomplishment promises key mechanistic, diagnostic and therapeutic insights. One major step in this direction is the identification of genes that drive the emergence of tumors upon mutation. Recent advances in the field of computational biology have shown the potential of combining genetic summary statistics that represent the mutational burden in genes with biological networks, such as protein–protein interaction networks, to identify cancer driver genes. Those approaches superimpose the summary statistics on the nodes in the network, followed by an unsupervised propagation of the node scores through the network. However, this unsupervised setting does not leverage any knowledge on well-established cancer genes, a potentially valuable resource to improve the identification of novel cancer drivers. Results We develop a novel node embedding that enables classification of cancer driver genes in a supervised setting. The embedding combines a representation of the mutation score distribution in a node’s local neighborhood with network propagation. We leverage the knowledge of well-established cancer driver genes to define a positive class, resulting in a partially labeled dataset, and develop a cross-validation scheme to enable supervised prediction. The proposed node embedding followed by a supervised classification improves the predictive performance compared with baseline methods and yields a set of promising genes that constitute candidates for further biological validation. Availability and implementation Code available at https://github.com/BorgwardtLab/MoProEmbeddings. Supplementary information Supplementary data are available at Bioinformatics online.

Novel Driver Strength Index highlights important cancer genes in TCGA PanCanAtlas patients

“…Some algorithms try to predict driver strength based on data from protein interaction networks (1)(2)(3). The idea is that a gene having multiple connections with other genes, i.e.…”

Linear functional organization of the omic embedding space

“…To systematically evaluate the predictive performance of our approach, we repeat the train/test process, for the protein complexes and the cancer-driver genes, 20 times and we compute the average percentage of correct predictions (accuracy). As a baseline for comparison with our function assignment strategy (membership in a protein complex, or being cancer-related) we use the SVM, a state-of-the-art binary classifier for vectorial data that has recently been used to predict cancer-related genes ( Gumpinger et al , 2020 ), or protein function, which is a multiclass classification problem in the embedding space ( Cho et al , 2016 ; Gligorijević et al , 2018 ). Similar to these approaches, in the case of the protein complexes, a multiclass classification problem, we are using the One-versus-One heuristic method to apply the SVM.…”

“…To the best of our knowledge, this is the first method that exploits directly the proximity in the embedding space by doing simple linear operations to identify cancer genes; recent studies (e.g. Gumpinger et al , 2020 ) trained a classifier with the vectorial representations of genes to identify cancer-related genes.…”

Machine learning methods for prediction of cancer driver genes: a survey paper

“…recognize patterns in a local neighborhood of a node, using the PPI network as the input graph. A similar approach was proposed by Gumpinger et al [56], using the InBio Map PPI network as the representation of gene interactions to generate node embeddings that integrate the network structure with nodes' MutSig p-values. Their findings suggest that a node's context within a network introduces valuable information in the prediction of cancer drivers.…”