Graduation Year

2009

Document Type

Dissertation

Degree

Ph.D.

Degree Granting Department

Biology (Cell Biology, Microbiology, Molecular Biology)

Major Professor

Gary Arendash, Ph.D.

Co-Major Professor

Professor: Huntington Potter, Ph.D.

Committee Member

Gordon Fox, Ph.D.

Committee Member

Patrick Bradshaw, Ph.D.

Committee Member

Chuanhai Cao, Ph.D.

Keywords

neuropathology, neural networks, caffeine, GRK5, GMCSF

Abstract

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder and the

leading cause of human senile dementia. Alzheimer’s represents a significant public

health concern, having widespread social and economic implications. Consequently,

protocols for early detection and therapeutic intervention (both behavioral and pharmacologic)

constitute important targets for medical investigation. Furthermore, contemporary

research depends upon comprehensive neurobehavioral assessment and

advanced statistical and computational analytic methodologies for characterizing

AD-associated sensorimotor and cognitive impairment, as well as evaluating therapeutic

efficacy. This dissertation introduces data mining-based techniques (decision

trees, neural networks, support vector machines) for behavioral analysis in both

nontransgenic and Alzheimer’s transgenic mice, to evaluate the cognitive benefits of

long-term caffeine treatment. Both treatment and transgenic effects are identified

through advanced statistical (discriminant analysis) and data mining approaches. In

addition, a novel mouse-based cognitive assessment paradigm, adapted from a human

interference learning AD-diagnostic protocol, is implemented to evaluate both

genetic (GRK5) and therapeutic (GM-CSF) effects in mice, against an Alzheimer’s

transgenic background. Data mining techniques are shown to be comparable to con

ventional statistical analyses, often providing complementary diagnostic information.

Indeed, comparisons between data mining-based and multivariate statistical analyses,

with respect to groupwise discriminability, support the use of both methodologies

in neurobehavioral research. Future work involving both data mining-based and

multivariate statistical analyses of cognitive-behavioral data is discussed, emphasizing

the need for longitudinal studies, repeated-measure designs, and spatiotemporal

modeling for evaluating the time-course of both human AD and AD-like pathology

in transgenic mouse models.

Share

COinS