Immune checkpoint inhibitors are immunotherapy drugs designed to help the immune system find and attack cancer cells within the body. They are often offered as adjuvant therapy, which is typically given after the primary treatment to lower the risk of the cancer returning, and have shown to improve survival rates.

Although the therapy has significantly improved treatment, many patients still experience resistance and side effects. Wanting to get a deeper understanding, Moffitt Cancer Center researchers developed a model called Auto-Encoder Survival Deep Network to predict the prognosis and response to therapy. This is similar to the Immunoscore method, which predicts the risk of relapse in people with colon cancer by measuring the patient’s immune response at the tumor site.

The analysis used artificial intelligence to identify prognostic biomarkers in patients treated with immune checkpoint inhibitors, utilizing a pan-cancer approach, or assessing frequently mutated genes and other genetic abnormalities common to many different cancers.

“We are using AI to investigate the relationship between the genes and the overall survival of patients,” said Payman Ghasemi Saghand, Ph.D., a research fellow at Moffitt.

The findings from the study were presented at the 2022 American Society of Clinical Oncology Annual Meeting.

"Using our developed models, we are doing several layers of analysis to study what are the genes that differentiate the patients with low risk and high risk and how these genes are connected."

- Dr. Payman Ghasemi Saghand, @GhasemiPayman

Within the study, researchers analyzed the RNA data of 522 patients who received immunotherapy, looking at six different types. Since immune checkpoint inhibitors are used to treat various types of cancers, for this study, the dataset included RNA of patients with four cancer types; melanoma, kidney, head and neck, and lung.

“Based on the RNA data of the patients and how long they survived after receiving immunotherapy, we tried to find a set of genes and the optimal combination of the gene expressions, to see which patients are a good fit for immunotherapy,” said Ghasemi Saghand.

The results showed that when using the Auto-Encoder Survival Deep Network analysis compared to the Immunoscore, it’s a promising approach to identifying relevant prognostic biomarkers in cancer patients treated with immune checkpoint inhibitors.

“Using our analysis, we were able to identify patients with significantly better survival chances,” said Ghasemi Saghand.

The researchers believe their findings may lead to novel therapeutic predictive signatures and identification of mechanisms of immune checkpoint inhibitors efficiency.

Ghasemi Saghand says the next steps are to continue working on more complicated approaches, performing analysis on each cancer category, analyzing to find the most important pathways, and comparing their findings with pathways reported in literature.

“We are currently working on novel and much more complicated artificial intelligence algorithms for this kind of analysis. Using our developed models, we are doing several layers of analysis to study what are the genes that differentiate the patients with low risk and high risk and how these genes are connected,” said Ghasemi Saghand.