by Anna Drozdz, Caitriona E. McInerney, Kevin M. Prise, Veronica J. Spence and Jose Sousa
Omics technologies are central to advancing the discovery of new cancer biomarkers and require highly sophisticated analysis due to the complexity of the data. The choice of analytical methods used can influence further biological inferences. In this study, we focused on discovering new genetic biomarkers for grade II and grade III astrocytoma brain tumours. Differentiating between these two grades is a significant challenge and is currently based solely on histopathological assessment, which is prone to inter-observer bias.
This study used feature selection and classification techniques to analyse astrocytoma samples and identified six distinct signature gene sets. Analysis of the impact of these sets on pattern discovery and classification revealed significant data noise and redundancy. Feature selection significantly narrowed the gene pool and improved classification accuracy, although gene selection was not consistent across methods, with minimal overlap between some. Discrepancies were found in both Gene Ontology and KEGG pathway analysis and their prognostic value. The study highlights the differences in results between pure classification algorithms and those that integrate feature selection. Additionally, the study showed how improperly designed analytical pipelines can lead to missed biological insights.
This work was carried out in collaboration with prof. Kevin Prise at the Queens Belfast University and supported by Brainwaves Northern Ireland and the Robin Menary Foundation for Brain Tumour Research.
Authors: Anna Drozdz, Caitriona E. McInerney, Kevin M. Prise, Veronica J. Spence and Jose Sousa
DOI: https://doi.org/10.3390/cancers16193263
Link to article: www.mdpi.com
Keywords: astrocytoma, genetics, breakthrough research, personal health