Thesis (MSc)--Stellenbosch University, 2014. / ENGLISH ABSTRACT: Background
New methods to facilitate exploratory analysis in scientific data are in high
demand. There is an abundance of available data used only for confirmatory
analysis from which new hypotheses can be drawn. To this end, two new
exploratory techniques are developed: one for chemometrics and another for
visualisation of fundamental scientific experiments. The former transforms
large-scale multiple raw HPLC/UV-vis data into a conserved set of putative
features - something not often attempted outside of Mass-Spectrometry. The
latter method ('StatNet'), applies network techniques to the results of designed
experiments to gain new perspective on variable relations.
Results
The resultant data format from un-targeted chemometric processing was
amenable to both chemical and statistical analysis. It proved to have integrity
when machine-learning techniques were applied to infer attributes of
the experimental set-up. The visualisation techniques were equally successful
in generating hypotheses, and were easily extendible to three different types
of experimental results.
Conclusion
The overall aim was to create useful tools for hypothesis generation in a
variety of data. This has been largely reached through a combination of novel
and existing techniques. It is hoped that the methods here presented are
further applied and developed. / AFRIKAANSE OPSOMMING: Agtergrond
Nuwe metodes om ondersoekende ontleding in wetenskaplike data te fasiliteer
is in groot aanvraag. Daar is 'n oorvloed van beskikbaar data wat slegs
gebruik word vir bevestigende ontleding waaruit nuwe hipoteses opgestel kan
word. Vir hierdie doel, word twee nuwe ondersoekende tegnieke ontwikkel: een
vir chemometrie en 'n ander vir die visualisering van fundamentele wetenskaplike
eksperimente. Die eersgenoemde transformeer grootskaalse veelvoudige
rou HPLC / UV-vis data in 'n bewaarde stel putatiewe funksies - iets wat
nie gereeld buite Massaspektrometrie aangepak word nie. Die laasgenoemde
metode ('StatNet') pas netwerktegnieke tot die resultate van ontwerpte eksperimente
toe om sodoende ân nuwe perspektief op veranderlike verhoudings te
verkry.
Resultate
Die gevolglike data formaat van die ongeteikende chemometriese verwerking
was in 'n formaat wat vatbaar is vir beide chemiese en statistiese analise. Daar
is bewys dat dit integriteit gehad het wanneer masjienleertegnieke toegepas
is om eienskappe van die eksperimentele opstelling af te lei. Die visualiseringtegnieke
was ewe suksesvol in die generering van hipoteses, en ook maklik
uitbreibaar na drie verskillende tipes eksperimentele resultate.
Samevatting
Die hoofdoel was om nuttige middele vir hipotese generasie in 'n verskeidenheid
van data te skep. Dit is grootliks bereik deur 'n kombinasie van oorspronklike
en bestaande tegnieke. Hopelik sal die metodes wat hier aangebied
is verder toegepas en ontwikkel word.
Identifer | oai:union.ndltd.org:netd.ac.za/oai:union.ndltd.org:sun/oai:scholar.sun.ac.za:10019.1/86366 |
Date | 04 1900 |
Creators | Emerton, Guy |
Contributors | Jacobson, Daniel, Stellenbosch University. Faculty of AgriSciences. Dept. of Institute for Wine Biotechnology. |
Publisher | Stellenbosch : Stellenbosch University |
Source Sets | South African National ETD Portal |
Language | en_ZA |
Detected Language | Unknown |
Type | Thesis |
Format | x, 95 p. : ill. |
Rights | Stellenbosch University |
Page generated in 0.0013 seconds