Multi-Dimensional Bayesian Classifier, Causal Polytree And Their Applications On HIV Mutation Pattern And Drug Resistance

Acquired Immune Deficiency Syndrome is also called HIV.Until now,there is still no methodology whose function fundamentally cures this epidemic,or prevents its spread.The ceaseless research indicates the hardness of curing HIV rests with the complex HIV mutation patterns,which go through unprecedented evolution under the drug pressure of HIV therapies.Since uncertain relationship between mutation patterns and drug resistance is still unknown,consequently,we are interested in using multi-dimensional Bayesian classifier and causal polytree to learn those causalities underline HIV mutation patterns,furthered with how those relationships influence HIV drug resistance.With enormous application success in expressing the complex and uncertain probabilistic dependencies in different situations,Bayesian network is a probabilistic graphical model that represents a set of variables and their probabilistic independencies.This thesis mainly focuses on analysis of multi-dimensional Bayesian classifier,and Bayesian causal polytree.With respect to the application fields,those models are directly applied to understand two concerned questions:1.What kind of relationships between mutation patterns? 2.How mutation patterns influence the drug resistance? In sum,my thesis contributes in two respects as follows.Our important theoretical contributions1 we analysis the methodology of training multidimensional Bayesian classifier and the complexity of inference problems MAP,MPE in multi-dimensional Ba- yesian classifier,together with the propose of training algorithm for mult i-dimensional Bayesian classifier based on score principle,and an efficient exact inf- erence algorithm for MPE based on gray-code searching space.2 We extend Pearl's orienting principles on Bayesian causal polytree,and solve the issues of minimal causal basins and unique polytree recover.We also prove approximate ratio of our designed algorithm for causal polytree training is K+1.Our important application contributions1 Based on HIV database with 5692 sequence provided by Stanford HIV research group,multi-dimensional Bayesian classifiers are trained to understand how HIV mutation patterns influence the drug resistance level;2 Based on our algorithm named Unique Polytree Recover,we train Bayesian causal polytrees for important mutations in therapies PIs,NRTIs,NNRTIs individually,the explanations under those models unbelievable coincide with conclusions of clinical experiments.Besides,those models estimate new characteristics of mutation patterns. 3 We develop a MATLAB toolbox including the important algorithms of Bayesian models and HIV sequence analysis,involving data abstraction, model learning,random data simulation,computation of inference and so forth.As for our open resource,please refer to our space in MATLAB website,http://www.mathworks.com/matlabcentral/fileexchange/authors/45984...