Harnessing the power of AI to learn more about deadly cancers

The AI combines data from multiple platforms to identify potential therapeutic targets.

Ayesha Gulzar
Harnessing the power of AI to learn more about deadly cancers
Deadly disease concept

da-kuk/iStock 

A team of scientists from the United States, France, and Italy have developed a robust machine-learning algorithm that uncovers potential targets for treating aggressive brain tumors, according to a study published on Feb .02.

The study focused on Glioblastoma multiforme (GBM), a fast-growing, aggressive, and usually fatal brain cancer originating in glial cells — cells that provide support and protection for the nerve cells. Glioblastoma is the most common and aggressive type of malignant brain tumor and often has a poor prognosis.

Machine learning strengthens a multi-decade effort

Although many potential targets for cancer treatment have been identified by studying proteins and genomes — proteogenomics — in human tumors, this information has not yet been proven helpful for personalizing cancer treatment.

Scientists have analyzed proteogenomic data through a machine learning algorithm called Substrate PHosphosite-based Inference for Network of KinaseS — SPHINKS — to identify the two most active kinases (enzymes), PKCδ and DNA-PK— termed as “master kinases.”

Patient-derived models revealed that PKCδ plays a crucial role in subtypes of glioblastoma characterized by metabolism and response to treatments targeting metabolic pathways.

DNA-PK plays a significant role in another subtype of glioblastoma characterized by its ability to rapidly produce new cells and develop into different types of cells. Therefore, targeting these proteins may be effective as a therapy for glioblastoma.

The developed algorithm generates a complete set of biological interactions to identify potent kinases that cause abnormal growth and treatment resistance in each glioblastoma subtype. As DNA information alone has proven insufficient in identifying tumor vulnerabilities and molecular mechanisms that drive each patient’s disease, this study may help bridge the gap by supplying more comprehensive data.

Patients can be classified based on common genes, proteins, fat molecules, epigenetics, metabolites, and other biological features.

How credible are the findings?

To validate patient-derived model findings, scientists used patient samples to grow their “tumor avatars” in the lab to demonstrate that drugs that target the activity of the master kinases can obliterate tumor growth.

Furthermore, the algorithm is mature enough to be integrated into any molecular pathology lab and has been provided for clinical use via a web portal. Omics information can be imported into the portal, and classification information can be obtained for multiple tumors, readily applicable to patient care.

Can the tool be used for other cancer types?

Although the machine learning tool has been developed for glioblastoma multiforme, it can also be used for breast, lung, and pediatric brain tumors. The same master kinases were found to be actionable therapeutic targets for these cancer types. The results are promising enough to motivate the team to conduct a new type of clinical trial.

Despite numerous drug therapies, glioblastoma prognosis has seen little improvement over the years, with survival rates below ten percent. However, the optimistic findings of this and similar studies may help doctors close the gaps towards better treatment and care, significantly impacting public health.

The study was a collaboration between researchers from the Institute for Cancer Genetics, Columbia University, and Sylvester Comprehensive Cancer Center, the University of Miami, Miller School of Medicine. The study was published in the Journal Nature Cancer.

Study abstract:

Despite producing a panoply of potential cancer-specific targets, the proteogenomic characterization of human tumors has yet to demonstrate value for precision cancer medicine. Integrative multi-omics using a machine-learning network identified master kinases responsible for effecting phenotypic hallmarks of functional glioblastoma subtypes. In subtype-matched patient-derived models, we validated PKCδ and DNA-PK as master kinases of glycolytic/plurimetabolic and proliferative/progenitor subtypes, respectively, and qualified the kinases as potent and actionable glioblastoma subtype-specific therapeutic targets. Glioblastoma subtypes were associated with clinical and radiomics features, orthogonally validated by proteomics, phospho-proteomics, metabolomics, lipidomics and acetylomics analyses, and recapitulated in pediatric glioma, breast and lung squamous cell carcinoma, including subtype specificity of PKCδ and DNA-PK activity. We developed a probabilistic classification tool that performs optimally with RNA from frozen and paraffin-embedded tissues, which can be used to evaluate the association of therapeutic response with glioblastoma subtypes and to inform patient selection in prospective clinical trials.