Machine Learning Used to Find Brain Tumor Cells That Mimic Healthy Neurons in Glioblastoma Patients [Study]

Researchers used machine learning to detect brain tumor cells that mimic healthy neurons to survive. The new approach may help find potential therapies to kill the resistant glioblastoma cells.

Machine Learning To Find Glioblastoma Cells That Mimic Health Neurons

Some cancers are difficult to treat due to cells that are highly skilled in resisting current treatment. Glioblastoma is among them, but researchers found a way to spot treatment-resistant glioblastoma cells in a new study.

In a new study, researchers at Sylvester Comprehensive Cancer Center at the University of Miami Miller School of Medicine and collaborating organizations assembled matched tumor samples from 123 individuals with glioblastoma at the time of diagnosis and after the disease returned following initial therapy, the research team created what would eventually become the biggest dataset of its kind.

Researchers identified significant alterations not previously observed in comparable cancer studies that looked at the tumors' genomes or transcriptomes, the collection of RNA molecules in cancer cells, by analyzing the proteomes and protein modifications in the samples.

They proved that glioblastoma cells were proliferative before therapy by examining cancer proteins and their alterations, particularly phosphorylation, which is a unique modification. In this condition, the cells devote their energy to self-replication.

Chemotherapies target cell activities in self-replication, so glioblastoma cells copy healthy neurons so they won't be discovered. The cells resemble normal neurons when the illness recurs.

With the new dataset, the scientists focused on kinases- enzymes that phosphorylate other proteins- to find possible treatments to eradicate these resistant tumors.

Migliozzi and colleagues identified the most active kinases in the neuron-like glioblastomas using a machine-learning technique they had previously developed. Kinases are essential for many biological processes and are major targets for several FDA-approved cancer treatments. Of the kinases, BRAF stood out.

In certain malignancies, such as melanoma, the gene encoding for this kinase is frequently altered; yet, in glioblastoma, BRAF protein levels rise without corresponding gene alterations. If the cancer proteome had not been examined, the researchers would not have made this significant discovery.

They next used a patient-derived xenograft tumor in mice and treatment-resistant glioblastoma cells in a petri dish to evaluate vemurafenib, an established BRAF inhibitor.

When the medication was used with the chemotherapy medicine temozolomide, the previously resistant tumors were successfully removed in both cases. The BRAF inhibitor increased the mice's lifespan in the model more than chemotherapy did.

Antonio Iavarone, M.D., deputy director at Sylvester, who led the study, believes their artificial intelligence method to identify glioblastoma's most active kinase can be used for other cancer types. By identifying the highest active kinase in each tumor and treating it with an already-approved kinase inhibitor, he and his researchers hope to create a clinical test using artificial intelligence (AI) to detect treatable vulnerabilities in a range of tumors.

Why Glioblastoma Cells Mimic Healthy Neurons?

Cancer treatments target cells that replicate as they typically grow faster than healthy cells. The researchers noticed that glioblastoma cells replicate neurons to evade treatments. The cancer cells want to live.

"The tumor cells actually resemble normal brain cells," said Simona Migliozzi, Ph.D., an assistant scientist at Sylvester and one of the study's lead authors. "Why? Because tumor cells want to survive, they want to live, and they're able to acquire therapy resistance by mimicking the normal brain."

Check out more news and information on Cancer in Science Times.