The Study Of Predicting Disease Genes Based On Complex Network

Identifying genes associated with disease has become a challenge work in life science field. Traditional methods of predicting disease genes include linkage analysis and association study. However, linkage analysis can only identify a chromosomal region where unknown disease genes are located, and this region usually contains tens to hundreds of genes. Association study also needs to clearly know candidate genes. So in recent years, the researchers in the world are successively proposing methods to further confirm disease genes in this region.With the completion of human genome project (HGP) and arise of high-throughput biological technology, we have acquired large-scale protein-protein interaction network (PPIs). Several studies have shown that two proteins having a higher topological overlap are more likely to belong to the same functional class than proteins having a lower topological overlap. Based on these evidences, we have put forward a method named MTOM&ATOM based on PPIs to predict disease genes. This method combine multi-nodes topology overlap measure (MTOM) and averaged topology overlap measure (ATOM).MTOM&ATOM measures the similarity of two subset of nodes through weighting the number of their neighborhood overlapping, which can better reflect biological significance. We test our method on 110 disease-gene families with total 783 genes, and leave-one-out cross validation show that our method can achieve enrichment up to 27-fold and area under roc curve (AUROC) are up to 92.3% on simulated linkage intervals of 100 genes surrounding the disease genes, which is better than others' results.Next, we apply MTOM&ATOM to the study of predicting genes related with Alzheimer's disease (AD). Firstly, based on PPIs which was constructed by kohler and others, we have achieved the same effect comparing to random walk(RW) measure. Secondly, based on brain specific network constructed by liu and others, among the top 46 high-scoring genes that we have found,40 were previously reported to be associated with AD, which is slightly better than liu's. The results show that our method has low complexity, high computing speed. And also it is robust to the incomplete of network and false interaction between two nodes. In our work, we also further confirm that the genes which have a high overlapping neighborhood with disease-causing genes are likely to cause either the same or similar diseases.