Show simple item record

dc.contributor.advisorBraga-Neto, Ulisses M
dc.creatorBahadorinejad, Arghavan
dc.date.accessioned2017-08-21T14:37:53Z
dc.date.available2017-08-21T14:37:53Z
dc.date.created2017-05
dc.date.issued2017-04-25
dc.date.submittedMay 2017
dc.identifier.urihttp://hdl.handle.net/1969.1/161434
dc.description.abstractIt is well known that the molecular basis of many diseases, particularly cancer, resides in the loss of regulatory power in critical genomic pathways due to DNA mutations. We propose a methodology for model-based fault detection and diagnosis for stochastic Boolean dynamical systems indirectly observed through a single time series of transcriptomic measurements using Next Generation Sequencing (NGS) data. The fault detection consists of an innovations filter followed by a fault certification step, and requires no knowledge about the system faults. The innovations filter uses the optimal Boolean state estimator, called the Boolean Kalman Filter (BKF). We propose an additional step of fault diagnosis based on a multiple model adaptive estimation (MMAE) method consisting of a bank of BKFs running in parallel. The efficacy of the proposed methodology is demonstrated via numerical experiments using a p53-MDM2 negative feedback loop Boolean network. The results indicate the proposed method is promising in monitoring biological changes at the transcriptomic level. Genomic applications in the life sciences experimented an explosive growth with the advent of high-throughput measurement technologies, which are capable of delivering fast and relatively inexpensive profiles of gene and protein activity on a genome-wide or proteome-wide scale. For the study of microbial classification, we propose a Bayesian method for the classification of r16S sequencing pro- files of bacterial abundancies, by using a Dirichlet-Multinomial-Poisson model for microbial community samples. The proposed approach is compared to the kernel SVM, Random Forest and MetaPhyl classification rules as a function of varying sample size, classification difficulty, using synthetic data and real data sets. The proposed Bayesian classifier clearly displays the best performance over different values of between and within class variances that defines the difficulty of the classification.
dc.format.mimetypeapplication/pdf
dc.language.isoen
dc.subjectFault detection
dc.subjectClassification
dc.titleFault Detection and Diagnosis in Gene Regulatory Networks and Optimal Bayesian Classification of Metagenomic Data
dc.typeThesis
thesis.degree.departmentElectrical and Computer Engineering
thesis.degree.disciplineElectrical Engineering
thesis.degree.grantorTexas A & M University
thesis.degree.nameDoctor of Philosophy
thesis.degree.levelDoctoral
dc.contributor.committeeMemberDougherty, Edward
dc.contributor.committeeMemberSerpedin, Erchin
dc.contributor.committeeMemberIvanov, Ivon
dc.type.materialtext
dc.date.updated2017-08-21T14:37:54Z
local.etdauthor.orcid0000-0001-6925-4090


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record