Big Data Powers Up Search for Immunotherapy Responders

Big Data Powers Up Search for Immunotherapy Responders
Biostatisticians introduce Cancer-Immu, the most comprehensive platform to date for identifying biomarkers for immune checkpoint blockage response.

A Vanderbilt University Medical Center team has developed a large and dynamic cancer atlas to help clinicians and researchers better predict their patients’ likely response to immune checkpoint blockade (ICB) therapy.

The team, led by Qi Liu, Ph.D., associate professor of biostatistics, and Yu Shyr, Ph.D., the Harold L. Moses Chair in Cancer Research, published its work in Cancer Research in February 2022. Jing Yang, Ph.D., a postdoctoral fellow in biostatistics, was first author.

“The goal of identifying responders has been limited by small sample sizes and ungeneralizable performance across cohorts,” Shyr said. “We set out to develop an atlas that overcomes these shortfalls through meta-analyses capabilities and a pan-cancer search engine.”

“Using Cancer-Immu, a user can input an individual patient’s genomic and other data and learn the strengths of associations with others who had the same cancer type and have been immunotherapy responders,” Yang said. “Or, if the cancer in question has sparse patient data, they can do searches across the 16 common cancer types to look for meaningful associations.”

Cancer-Immu supplies the needed power by combining data on over 3,652 patients (at the time of publication) across all the major cancers. As the database grows, and as ongoing survival data are added, clarity on predictive biomarkers, like gene variants or mutations, will emerge.

“We hope we find a strong predictive biomarker for each immunotherapy drug across all cancers,” Shyr said. “That will be the dream, to know in advance which patients should be given immunotherapy, and truly provide that precision-medicine approach.”

Increasing Study Power

ICB emerged as a breakthrough cancer treatment in the early 2000s, boosting response to just shy of “cured” in many patients. While this has been famously true for formerly intractable cancers like metastatic melanoma (with about a 50 percent response rate) and non-small cell lung cancer (with about a 20 percent response rate), Shyr says it is also proving true to varying degrees for more than a fifth of cancer patients across the spectrum of common cancer types.

“The goal of identifying responders has been limited by small sample sizes and ungeneralizable performance across cohorts. We set out to develop an atlas that overcomes these shortfalls through meta-analyses capabilities and a pan-cancer search engine.”

Knowing who will respond has profound consequences on opportunities seized or missed. A delay or a wrong turn – whether relying on ICB at the expense of chemotherapy or the converse – could be a disastrous decision for the patient.

Unfortunately, little is known in advance about which cases are likely to respond. Parsing the responders and non-responders by broad categories like age, race and cancer stage has failed to uncover commonalities.

Researchers have come to understand that it will take a granular examination of a patient’s profile, probably along many parameters, to identify the biomarkers that predict ICB response. The path to this lies in gathering and analyzing enough detailed data on enough patients, across enough cancers, that the biomarkers begin to emerge in a statistically meaningful way.

Options for the Cancer-Immu User

Cancer-Immu is used to prioritize multiomics features associated with ICB response. These features include DNA sequence data and RNA expression data (both bulk and single-cell), along with patient clinical outcome and cancer response: whether the tumor grew, remained unchanged, shrank, or disappeared.

“That will be the dream, to know in advance which patients should be given immunotherapy, and truly provide that precision-medicine approach.”

Since needs vary, the atlas draws on two powerful analytical tools: a meta-analysis that combines results from multiple studies; and a pan-cancer analysis, a pooling of multiple datasets from different cancers into one large dataset to search for associations.

“The meta-analysis reveals consistent signatures across multiple study cohorts, while pan-cancer analysis enhances our ability to detect and analyze rare features by aggregating samples across both cohorts and tumor types,” Yang said.

Examiners have the option of uploading and analyzing a data set independently or co-analyzing their data along with the existing database.

Power from More Data and AI

Shyr’s current mission is to strengthen Cancer-Immu’s power by continuously increasing sample size (number of cases) and adding follow-up survival data as it accumulates.

“Most importantly we’re going to bring more comprehensive advanced machine learning into the system,” Shyr said. “I liken it to this: instead of describing a cat as a furry animal with pointy ears, the system will see cats in every which way – in shadow and in sunshine, from the side and from the front – and it will learn what to look for much more quickly. This will provide shortcuts for anyone trying to extract meaningful associations.”