TarGo: network based target gene selection system for human disease related mouse models

Abstract

Genetically engineered mouse models are used in high-throughput phenotyping screens to understand genotype-phenotype associations and their relevance to human diseases. However, not all mutant mouse lines with detectable phenotypes are associated with human diseases. Here, we propose the "Target gene selection system for Genetically engineered mouse models" (TarGo). Using a combination of human disease descriptions, network topology, and genotype-phenotype correlations, novel genes that are potentially related to human diseases are suggested. We constructed a gene interaction network using protein-protein interactions, molecular pathways, and co-expression data. Several repositories for human disease signatures were used to obtain information on human disease-related genes. We calculated disease- or phenotype-specific gene ranks using network topology and disease signatures. In conclusion, TarGo provides many novel features for gene function prediction.

Keywords: Bioinformatics; Database; Genetic engineered mice; PageRank algorithm; Systems biology.

Fig. 1

Prediction of gene-phenotype association using interaction network and signature genes. The hyperlink matrix and dangling matrix (no outgoing edge matrix) were constructed from interaction databases (red box). The weight matrix was constructed using public annotation databases (orange box). The d represented the dumping factor (0.85). The TSPR score represented the gene association for that particular phenotype. For this given TSPR score vector, A and D were signature genes. The seed nodes were selected from the top three ranking TSPR scores. A and D were good seeds because these two genes were signature genes in the input phenotype and ranked among the top three in the TSPR result. Therefore, d is 2 in this figure, meaning the good seed vector is ½. All other cases were given 0. Finally, phenotype-associated genes were selected from those with a high TrustRank score and low Spammass score

Fig. 2

ROC curve for TarGo prediction results. a In the MP term, the AUC score of TSPR is 0.66, TrustRank is 0.82, and SM filter is 0.91. b In the MeSH term, the AUC score of TSPR is 0.63, TrustRank is 0.95, and SM filter is 0.93. c Overlapping between IMPC and TarGo. the X-axis indicates high-ranking genes sorted by rank score. The Y-axis is the overlapping percentage between IMPC and TarGo

Fig. 3

Contents of the TarGo database. a The user can find the knock out (KO) target gene using the TarGo search system. Known phenotype, KO mouse state, and high-related phenotype are provided using gene name (blue line). High-associated genes are provided using multiple phenotype selection (red line). b The user can analyze gene rank using in-house data. The researcher can also use various in-house data, including gene sets from the Omics approach or empirical knowledge. Therefore, the user can regenerate gene rank with user-defined gene sets or network (green line)