Application Of Multidimensional Scaling Analysis Method To Study The Evolution And Spread Of H5N1Avian Influenza Virus Mechanism

This dissertation visualizes the dynamic evolution of the H5N1virus withmultidimensional scaling (MDS) algorithm, and further studies the global migrationmechanism of H5N1avian influenza.The massive growth of genetic data brings with bioinformatics, not only anopportunity but also a challenge. MDS can convert genetic distances to Cartesiancoordinates for the purpose of clustering and visualization. Classical MDS is notsuitable for analysis of massive biological sequence data, because of its highcomputational complexity. This dissertation analyzes classical MDS and its variations,and compares those algorithms in terms of numerical parallelism. Based on this,LMDS and SC-MDS are chosen as study objects for parallelization. Because the mosttime-consuming part in MDS algorithms is to calculate the biggest p eigenvalues andtheir (orthonormal) eigenvectors, the dissertation analyzes the numerical solution ofeigenvalue of symmetric real matrix, and chooses singular value decomposition(SVD) algorithm as a solution for MDS parallelization using CUDA. The laterefficiency test shows great speedup effect, illustrating16~17and84~218timesspeedup for LMDS and SC-MDS respectively. In order to reduce or stabilize randomerror, this thesis presents a global distribution based position method and validates itseffectiveness.Highly pathogenic avian influenza (HPAI) H5N1virus is a non-eradicablezoonosis that continues to mutate and reassort in nature, and still poses a seriousthreat to avian and human health. With the parallelized MDS algorithm, thisdissertation reduces quickly the dimensions of H5N1sequence data and demonstratesthe dynamic evolution of the H5N1virus, providing another way of visualizationother than evolutionary tree.Understanding the global migration dynamics of highly pathogenic avianinfluenza HPAI H5N1viruses is important in surveillance and disease prevention.This dissertation estimates genetic diversity of H5N1virus isolated from differentregions and migration rates between regions. Based on this, it characterizes the globalmigration network and studies the migratory mechanism of HPAI H5N1viruses. Thegenetic analysis of HPAI H5N1viruses in this dissertation supports a globalpersistence model, in which each region acts to some extent as a source, and the migration network approximates the major flyways of migratory birds. The H5N1genealogy notably contains long side branches, with some lineages persisting foryears. This feature supports a model of local persistence of HPAI H5N1viruses.