Discoverig Genomic Patterns Using Fuzzy Self Organising Maps
Loading...
Date
2009-02
Authors
Chun, Ho Yi
Journal Title
Journal ISSN
Volume Title
Publisher
Universiti Sains Malaysia
Abstract
Microarray technology has allowed for the collection of thousands of genetic data
simultaneously. The problem with microarray data lies in the fact that while there
are thousands of attributes, the number of samples is few. Clustering becomes a
practical option in analysing such data as it is impractical to study every data
individually. Although there are numerous clustering methods such as k-means and
hierarchical clustering, it is insufficient when it comes to a massive scale dataset.
Self-organising maps (SOM) is a concept introduced by Teuvo Kohonen that
allowed for an unsupervised clustering and subsequent visualisation of data. This
visualisation is by means of projecting the data onto a 2-dimensional map grid while
retaining relations among map units that represent the data. Furthermore, the use of
component planes in SOM allows for the analysis of individual attributes and
correlation between attributes. On the whole, this clustering and visualisation allows
for a “big picture” analysis of a massive dataset. In genetics, hard clustering may not
be entirely suitable as multiple genes can have multiple traits, and thus, in clustering,
can belong to multiple clusters. This brings up the subject of incorporating fuzzy
rules, where one attribute is not restricted to have only one characteristic. Therefore,
a feasible solution would be a hybridisation of fuzzy c-means and SOM. While
fuzzy SOM (FSOM) has been explored, it requires many parameters prior to
processing. This work proposes a modified FSOM that requires very few initial
parameters, and was tested on a brain tumour dataset and a breast cancer dataset.
The results showed that the modified FSOM performed better than the ordinary
SOM, and the results of the cluster analysis matched a prior work.
Description
Keywords
Genomic Patterns , Fuzzy Self Organising Maps