Authors:
Maysson Al-Haj Ibrahim
1
;
Joanne L Selway
1
;
Kian Chin
2
;
Sabah Jassim
1
;
Michael A. Cawthorne
1
and
Kenneth Langlands
1
Affiliations:
1
Buckingham University, United Kingdom
;
2
Milton Keynes Hospital NHS Foundation Trust, United Kingdom
Keyword(s):
Disease Classification, Breast Cancer, Prognosis, Biomarkers, Metabolic Networks and Pathways, Gene
Regulatory Networks, Microarray Analysis.
Related
Ontology
Subjects/Areas/Topics:
Algorithms and Software Tools
;
Bioinformatics
;
Biomedical Engineering
;
Pattern Recognition, Clustering and Classification
;
Transcriptomics
Abstract:
Accurate prognostication is central to the management of breast cancer, and traditional clinical and
histochemical-based assessments are increasingly augmented by genetic tests. In particular, the use of
microarray data has allowed the creation of molecular disease signatures for the early identification of
individuals at elevated risk of relapse. However, tailoring therapy on the basis of a molecular assay is only
recommended in certain cases, and the identification of a minimal set of genes whose expression allows
informed decision-making in a broader spectrum of disease remains challenging. Finding an optimal
solution is, however, an intractable computational task (i.e. retrieving the smallest group of genes with the
greatest prognostic power). Our solution was to reduce the genetic search-space by using two filtering steps
that enriched by biological function those genes whose expression discriminated disease states. In this way,
we were able to identify a new molecular signature
, the expression characteristics of which facilitated the
classification of intermediate risk disease. We went on to create a statistical test that confirmed the
relevance of our approach by comparing the performance of our signature to that of 1000 random
signatures.
(More)