Somatic mutation accumulation is a major cause of abnormal cell growth. However, some mutations in cancer cells may be deleterious to the survival and proliferation of the cancer cells, thus offering a protective effect to the patients. We investigated this hypothesis via a unique analysis of the clinical and somatic mutation datasets of ovarian carcinomas published by the Cancer Genome Atlas. We defined and screened 562 macro mutation signatures (MMSs) for their associations with the overall survival of 320 ovarian cancer patients. Each MMS measures the number of mutations present on the member genes (except for TP53) covered by a specific Gene Ontology (GO) term in each tumor. We found that somatic mutations favorable to the patient survival are predominant in ovarian carcinomas compared to those indicating poor clinical outcomes. Specially, we identified 19 (3) predictive MMSs that are, usually by a nonlinear dose-dependent effect, associated with good (poor) patient survival. The false discovery rate for the 19 "positive" predictors is at the level of 0.15. The GO terms corresponding to these MMSs include "lysosomal membrane" and "response to hypoxia", each of which is relevant to the progression and therapy of cancer. Using these MMSs as features, we established a classification tree model which can effectively partition the training samples into three prognosis groups regarding the survival time. We validated this model on an independent dataset of the same disease (Log-rank p-value < 2.3 × 10(-4)) and a dataset of breast cancer (Log-rank p-value < 9.3 × 10(-3)). We compared the GO terms corresponding to these MMSs and those enriched with expression-based predictive genes. The analysis showed that the GO term pairs with large similarity are mainly pertinent to the proteins located on the cell organelles responsible for material transport and waste disposal, suggesting the crucial role of these proteins in cancer mortality.