Artificial intelligence and brain cancer: new mapping to improve diagnosis and management of glioblastoma

While glioblastoma is a highly aggressive brain tumor that currently offers little hope of cure, researchers at the CANTHER laboratory (CNRS / Inserm / University of Lille / Lille University Hospital / Pasteur Institute of Lille) and the Laboratoire de Bioimagerie et Pathologies (CNRS / University of Strasbourg) have succeeded in identifying different forms of this tumor and mapping them precisely by analyzing the activity of gene regulatory factors. This new data paves the way for better management of this cancer and the development of new, more targeted treatments.

Glioblastoma is the most common brain tumor, but also the most aggressive. Every year, around 3,500 new cases are diagnosed in France. Today, despite numerous scientific advances, this tumor remains incurable due to its high molecular and cellular heterogeneity, which complicates the use of standard therapeutic regimens.

"The problem is that each tumor is unique: the genes expressed are numerous and different, forming a complex network of interactions. Our work reveals a hierarchy controlled by ’master regulators’ - hyper-connected key molecules - which actively maintain the tumor," explains Mohamed Elati, head of the "Digital Systems & Cancer Computational" team at the CANTHER laboratory in Lille. Seeking to categorize tumors in order to refine treatments, the scientists had so far succeeded in identifying 4 tumor subgroups based on patients’ transcriptional profiles (gene expression). But some groups were still very heterogeneous.

In this new study, researchers focused on the activity of regulatory molecules - transcription factors - which interact with our genes, activating or inhibiting their expression. Of the 2375 transcription factors and cofactors present in humans, 539 are active in glioblastoma tumor mechanisms. Thanks to AI and machine learning in particular, the researchers were able to unify data from 16 international studies conducted over several years (i.e. around 1,600 patients). This approach has enabled them to establish the largest mapping to date of gliobastoma transcriptional activity, this time identifying no fewer than 7 tumor subtypes, each associated with specific biological mechanisms and a different prognosis (see figure).

Transcription factor activity mapping of the 1600 cases identified. Each color is associated with a specific tumor type with specific mechanisms.

This bioinformatics tool, made available to the scientific community and named GBM-cRegMap, aims to determine precisely, on the basis of individual molecular data, the characteristics of the tumor at the time of detection, but also after treatment at the time of recurrence. These data are invaluable for a better understanding of the mechanisms of glioblastoma and its evolution, and for the development of more personalized therapies.

The map also reveals that current preclinical models (cell models simulating the tumor and enabling new therapies to be tested) do not, in reality, respond to all the tumor types identified, underlining the need to develop new cell lines.