Mobile Menu

New study uses machine learning and RNA-seq to identify cancer-causing gene fusions in glioblastoma

Written by Miyako Rogers, Science Writer

A new study, published in Nature, has identified gene fusions with oncogenic potential in glioblastoma. Using Oncofuse, a machine learning tool, researchers interrogated RNA-seq data from the tumours of 139 newly diagnosed glioblastoma patients. 30 fusions were identified as having predictable oncogenic potential.  This study marks a good starting point for elucidating which gene fusions may affect the pathogenesis of glioblastoma, as well as other cancers.

Gene fusions

Glioblastoma is the most aggressive primary brain tumour, with an average survival length of around 14-16 months. Gene fusions, which are present in all cancerous tumours, may provide new insights for treatment strategies, and also might act as powerful prognostic biomarkers.

Gene fusions are chimeras: they can either be the result of genomic rearrangements that give rise to a single transcription unit or come about due to trans-splicing. RNA-seq can identify gene fusions easily; and whilst some gene fusions have been shown to play a role in the early stages of tumour development, many do not appear to have any functional consequences. As a result, working out which gene fusions are functionally relevant requires further analysis, and this study used a machine learning tool, Oncofuse, to do just that.

Gene fusions with oncogenic potential

This study used RNA-seq data from tumours of 139 newly diagnosed glioblastoma patients, from multiple databases. They then performed in silico analysis of the fusions detected, and found 103 different fusions, involving 167 different genes. They then eliminated some of these fusions by only selecting in-frame fusions that were able to produce a product. After that, they ran Oncofuse, and whittled down the number of fusions to just 30. These 30 fusions with predictable oncogenic potential were then classed into 4 distinct categories: 6 were previously described in cancer, 6 involved oncogenes or tumour-suppressor genes, 4 were predicted by Oncofuse to have oncogenic potential, and 14 other in-frame fusions. Only 2 of these fusions were present in more than one patient: FGFR3::TACC3 and EGFR::SEPTIN14.

Future implications

This study is really a starting point for future investigations. The gene fusions identified are rare, only detected in <1% of cases. However, characterising the genetic alterations of tumours, even in only a small subset of cases, will help identify specific tumour biomarkers for glioblastoma. This method of very specific analysis may elucidate differences between cancers, as well as inform precision medicine approaches for specific types of cancer.


More on these topics

Genomics / Machine Learning / Multi-omics

Share this article