Features Analysis And Prediction Of Aging Genes And Cancer Genes

Aging and cancer are closely related: the incidence of cancer increases progressively with aging. Looking for aging genes and cancer genes is the foundation for understanding their relationships and also an important work in the area currently. Since the high cost of identifying aging genes and cancer gene in wet lab, it is important to develop algorithm to predict novel candidates from throughput data to free biologists from the boring gene screening work. Here, we for the first time systematic analyzed features of aging genes and developed an algorithm to predict novel ones. Some of the predictions have been validated by wet-lab experiments. To identify cancer genes from somatic mutations in recently published cancer genome, we developed a simple yet powerful algorithm with significantly better performance than the previous. Considering the limited functional knowledge of cancer genes as a major hurdle for cancer research, we developed an algorithm to predict their finer functions. And finally, based on protein interaction network, we analyzed features of human aging genes and cancer genes and some common ones have been found which give clues to understanding the relationships of aging and cancer.The main contributions are as follows:1. We systematically analyzed properties of aging genes and found that, when compared to genes not yet known to be involved in lifespan regulation, known aging genes display the following features: (1) longer genomic sequences and protein sequences, (2) a stronger tendency to co-express with other genes, (3) significant enrichment in certain functions and RNAi phenotypes, (4) higher sequence conservation, and (5) higher in several network topological features such as degrees in a functional interaction network. Based on these features, a support vector machine based algorithm was developed to predict novel aging genes. Using this algorithm, 243 new aging genes were predicted with precision higher than 0.85. More, the contribution of each feature to the prediction result was evaluated and we found that the functions enriched by aging genes contributed the most and the phonotypes the second.2. For the predicted longevity genes, we want to validate at least some of them through literatures or wet lab experiments. Literatures supporting some of the predictions have been found. For example, consistent with our prediction, vps-34 has been shown necessary for the lifespan extension of dietary restriction in C. Kenyon's lab. For wet-lab experiments, seven genes were chosen and knocked down for lifespan assay and we found that RNAi of B0025.1 or F58F12.1 extended while RNAi of F54C9.1 shortened, the lifespan of daf-2 animals.3. To identify cancer genes from somatic mutations in recently published cancer genome, we proposed an algorithm to identify novel candidates by cancer functions. Combining the function knowledge and number of non-silent mutations of genes, our algorithm performs significantly better than the previous one which is based on selection pressure and number of non-silent mutations. Finally, a list of 46 kinase genes, were suggested as candidates.4. The functional knowledge of cancer proteins and cancer pathways is limited, remaining as a major hurdle to cancer studies. Specifically, many cancer proteins are only annotated to high-level general GO categories. Here, we developed an efficient algorithm to find finer functions of the cancer proteins in a function-specific sub-netwok. By exploiting their previously known functions, 193 cancer proteins were predicted to finer functions. Furthermore, we selected a group of specific functions significantly enriched with known cancer proteins as cancer functions. Using the algorithm, 221 proteins were predicted to them, improving the connection of the function-specific interaction sub-networks and thus delineating cancer functions more integrally and clearly.5. To understand reasons that aging and cancer are closely related, features of aging genes and cancer genes were investigated and we found that they (1) are highly overlapped; (2) have similar network topological features showing both tend to locate centrally in the network; and (3) are inter and intra connected by many direct protein interactions. In summary, after multi-dimensional features of aging genes and cancer genes analyzed, three efficient algorithms have been developed to predict new aging genes, cancer genes and cancer genes'finer functions. Some of the predictions have been validated by wet-lab experiment. More, common features of aging genes and cancer genes have also been found. Our work will give a push to understanding the mechanisms of aging and cancer.