Artificial intelligence informs solutions for gastric cancer challenges

June 09, 2022

Mayo Clinic researchers have found a 32-gene signature that is a promising prognostic and predictive biomarker for patients with gastric cancer. The study was published in the February 2022 edition of Nature Communications.

Tae Hyun Hwang, Ph.D., a cancer researcher in the field of artificial intelligence (AI) and informatics at Mayo Clinic in Jacksonville, Florida, worked with a team of other researchers to develop and implement machine learning and AI algorithms that can help solve clinically relevant gastric cancer challenges.

"If this is a clinically relevant and biologically important problem, and if we can solve that problem, that brings impact," said Dr. Hwang. "The most important thing is how we figure out what is the most important problem, and then learning if there are any biological mechanisms we can solve using computational approaches."

For gastric cancer specifically, most patients are treated with surgery and chemotherapy depending on their disease staging. Models like this, Dr. Hwang explains, help further stratify patients to bring more precision care. They can be predictive of treatment response, create patient subgroups based on disease characteristics, and ultimately give physicians more guiding information as they make decisions about gastric cancer.

The study analyzed the somatic mutation profiles of 6,681 patients from 19 different cancer types. The data were collected from the Cancer Genome Atlas, a combined cancer genomics effort between the National Cancer Institute (NCI) and the National Human Genome Research Institute, and input into a machine learning algorithm.

NTriPath identifies prognostic molecular pathways

NTriPath is a machine learning algorithm previously discovered by Dr. Hwang and his colleagues. The algorithm integrates pan-cancer somatic mutation data, gene-to-gene interaction networks and pathway databases to identify prognostic cancer-associated molecular pathways. It's been used to identify signatures for renal cell carcinoma, bladder carcinoma, head and neck squamous cell carcinoma, and melanoma.

Using NTriPath, the researchers identified important pathways specific to gastric adenocarcinoma and defined four molecular subtypes for the disease. They were then able to test the clinical relevance of these subtypes and create a risk scoring model to predict both overall survival and response to therapies including chemotherapy and immune checkpoint blockade.

"This field is moving very quickly," said Dr. Hwang. "If we have one singular code, we can bring more effective therapeutics to patients."

Dr. Hwang explains that using AI in this way can standardize diagnosis and identify eligibility for emerging therapies for gastric cancer, such as immunotherapy, based on the likelihood that a patient will respond well to the treatment.

Findings and clinical relevance

Dr. Hwang and his team generated microarray-based mRNA expression profiles from pre-treatment tumor samples from 567 patients, 89% of which had stage 2 or 3 disease. All the patients had undergone surgery at Severance Hospital, Yonsei University College of Medicine in South Korea.

Previously, researchers found that the top three pathways identified by NTriPath yielded the best results. For gastric cancer, the top three pathways identified consisted of 32 genes including TP53, BRCA1, MSH6, PARP1 and ACTA2.

When NTriPath was applied, researchers found four distinct molecular subtypes with treatment response:

  • Group 1 overexpressed genes associated with the cell cycle and DNA repair.
  • Group 2 did not show a distinct pattern of overexpressed genes.
  • Group 3 overexpressed genes found in apoptosis signaling and cell proliferation pathways.
  • Group 4 overexpressed genes found in TGF-β, SMAD, estrogen signaling and mesenchymal morphogenesis pathways.

Group 1 was the subgroup with the best prognosis, but this group showed adverse effects on chemotherapy. This group, as well as group 3, was sensitive to immunotherapy. Group 3 was a poor prognosis subgroup, but patients who received chemotherapy showed prolonged survival. Group 4 was the poorest prognosis subgroup and had no response to chemotherapy or immunotherapy.

"A very small fraction of tumors actually respond to immunotherapy," said Dr. Hwang. "Our data shows that groups 1 and 3 do respond." Dr. Hwang explains that patients in those groups aren't necessarily microsatellite-stable (MSS) patients and would likely have been prescribed a different treatment path.

The researchers analyzed these results against previous studies to determine that the signature was not just restating existing classification systems. Dr. Hwang and his team believe this type of work should be validated using large cohorts of patient samples in a prospective clinical study. For that reason, they've selected one of the 32 genes to begin further testing in a prospective clinical study.

For more information

Cheong J, et al. Development and validation of a prognostic and predictive 32-gene signature for gastric cancer. Nature Communications. 2022;13:774.

Refer a patient to Mayo Clinic.