Identifying functionally important missense mutations in cancer by dynamics-based analysis and predicting pathogenicity/disease category of missense mutations

Missense mutations have various effects on protein structures, also leading to distorted protein dynamics that plausibly affects the function. We hypothesized that missense mutations in cancer-related genes selectively target hinge-neighboring residues that orchestrate collective structural dynamics. To test our hypothesis, we selected 69 cancer-related genes from the Cancer Gene Census (CGC) database and their representative protein structures from the Protein Data Bank. We first identified the hinge residues in two global modes of motion by applying the Gaussian Network Model. We then showed that missense mutations are significantly enriched on hinge-neighboring residues in oncogenes and tumor suppressor genes. We observed that several oncogenes (e.g., MAP2K1, PTPN11, and KRAS) and tumor suppressor genes (e.g., EZH2, CDKN2C, and RHOA) strongly exhibit this phenomenon. Next, we developed a computational pipeline to detect significantly enriched three-dimensional (3D) clustering of missense mutations around hinge residues by using the the Cancer Genome Atlas (TCGA) dataset. The hinge residues were also detected by applying a Gaussian network model for the modes 1 to 5. By systematically analyzing the PanCancer compendium of somatic missense mutations in nearly 10,000 tumors from TCGA, we identified candidate genes and mutations in addition to well known ones. For instance, we found significantly enriched 3D clustering of missense mutations in known cancer genes including CDK4, CDKN2A, TCL1A, and MAPK1. Besides these known genes, we also identified significantly enriched 3D clustering of missense mutations around hinge residues in PLA2G4A, which may lead to excessive phosphorylation of the extracellular signal-regulated kinases. Our results show that the consideration of clustering around hinge residues can help us explain the functional role of the mutations in known cancer genes and identify candidate genes. Furthermore, we proposed new features, named hinge-based, for pathogenicity prediction for missense mutations and show that hinge-based features improve pathogenicity prediction. Pathogenicity prediction of human missense variants remains a challenging problem. Existing computational models are basically binary classifiers predicting whether given missense variants are deleterious or neutral. We demonstrated a multilabel classification method that predicts not only the pathogenicity but also the disease category type of given missense variants. Moreover, existing computational models are based on sequence-, structural-, or protein dynamics-based analysis. We also showed that network topological properties of proteins significantly improve determining the pathogenicity of missense variants. We trained and tested our model PathDis with 20,361 missense variants. Then, we benchmarked by the area under the ROC curve (AUROC) evaluation metric score with a well-established prediction model which uses the same dataset. We observed that our model PathDis improves AUROC by 3%. Then, we tested PathDis with a different dataset. Also benchmarking based on this different dataset against other well-established prediction models demonstrated that PathDis' AUROC score is approximately 3% higher than the second highest AUROC score. In addition to high pathogenicity prediction results, PathDis has approximately 79% accuracy for predicting the disease category types (i.e., No Disease, Cancer, and Non-Cancer). Besides introducing a prediction model, we also characterized the missense variants by our sequence-, structure-, dynamics-, and network-based features for the disease category types. We observed that sequence-based, network-based, and structure-/dynamics-based features characterize No Disease, Cancer, and Non-Cancer missense variants, respectively.