Gene-Phenotypic Link Prediction Based On Heterogeneous Biomolecular Network

The phenotype of an organism is mainly formed by environmental factors and genetic differences.With the development of bioinformatics,people pay more and more attention to the relationship between biomolecules and biological phenotypes,but the biomolecular level data is very large,relying solely on biological experiments to study,costly and long cycle.Therefore,using the advanced algorithms in the computer field,the correlation prediction of some gene phenotypes through the heterogeneous network constructed by biomolecules has become a research hotspot.There are many networks of different levels and different organizational forms between biomolecules,such as gene transcriptional regulatory networks,protein interaction networks,and so on.In addition to the molecular network in vivo,there are many related networks for the influence of biomolecules on phenotypes,such as the network of genes and diseases.Our research is mainly based on the use of various correlation networks between biomolecules and phenotypes for association prediction research.The first research work in this paper collected data on the traits of genes and traits associated with large yellow croakers and their associations through existing research.Based on the associated data,an improved KATZ algorithm was proposed to predict the association between genes and trait phenotypes,and the results were verified.In the second work,we propose a meta-path based disease network capture algorithm(mpDisNet).The algorithm is based on genes,miRNAs,and heterogeneous biomolecular networks of disease structures.The selected meta-paths are used to randomly walk around the heterogeneous network to generate random walk instances of the disease,and then through the heterogeneous network’s Skip-Gram algorithm.A multidimensional vector of each disease was obtained,and then the cosine similarity was used to calculate the association between disease phenotypes,and the obtained correlation results were verified and analyzed.Experiments have proved that this method is superior to the traditional miRNA-overlap method.In addition,pathophysiological pathways associated with disease phenotypes have also been analyzed.