Use this URL to cite or link to this record in EThOS:
Title: Artificial neural network techniques to investigate potential interactions between biomarkers
Author: Lemetre, C.
Awarding Body: Nottingham Trent University
Current Institution: Nottingham Trent University
Date of Award: 2010
Availability of Full Text:
Access from EThOS:
Access from Institution:
High-throughput technologies in biomedical sciences, including gene microarrays, supposed to revolutionise the post-genomic era, have barely met the great expectations they inspired to the biomedical community at first. Current efforts are still focused toward improving the technology, its reproducibility and accuracy. In the meantime, computational techniques for the analysis of the data from these technologies have achieved great progresses and show encouraging results. New approaches have been developed to extract relevant information out from these results. However, important work needs to be further conducted in order to extract even more meaningful and relevant information. These techniques offer great possibilities to explore the overall dynamic held within a living organism. The potential information contained in their output can reveal important leads at deciphering the interconnection, interaction or regulation influences that can exist between several molecules. In front of an increasing interest of the scientific community toward the exploration of these dynamics, some groups have started to develop solutions based on different technologies to extract these information related to interactions. Here we present an Artificial Neural Network-based methodology for the study of interactions in gene transcriptomic data. This will be applied and validated in a breast cancer context. This manuscript will discuss the methodological optimisation to identify biomarkers of interest from high-throughput transcriptomic technologies; and it will show how the algorithms were brought forward to identify the potential relationship that may exist between the markers identified. It will illustrate and highlight the robustness of the methods by discussing some examples of application in different breast cancer studies. The present thesis will show that despite the great difficulty to obtain gold validation to prove the robustness of the approach; it has been possible to identify some relevant features able to highlight the promises held by this preliminary development of the method. The results obtained by trying to identify the correlated component within an artificial dataset suggest some interesting ability of the approach. Additionally, when applied to the van’t Veer dataset (van’t Veer et al., 2002), the list of selected transcripts held two different isoforms for two different genes, and the method identified the strong correlation between the 2 forms. Finally, the results involving the transcripts for DTL, TK1 and CDC45L have been shown to overlap with the result of a similar work from Gevaert et al. (2006) on the van’t Veer dataset using a different method involving a Bayesian network with Markov blanket. Ultimately, this thesis will try to discuss the advantages or limitations as well as the potential application and future hopes around the methods introduced.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available