Rudolf Fehrmann

University of Groningen - University Medical Center Groningen

Using machine learning to unravel the 'cancer-immune setpoint' in rare cancers (SETPOINT)

During cancer development, tumor cells undergo molecular ‘rewiring’ to escape the immune system, often by activating mechanisms that suppress an anti-cancer immune response. Immunotherapy can circumvent some of these mechanisms and thus trigger an anti-cancer response, however, little is known about the efficacy of immunotherapy in patients with rare cancers. Most research is being done in a limited set of common cancer types and little is known about the factors that influence the anti-cancer immune setpoint in rare cancers. To improve our understanding of the mechanisms that determine the triggering of an anti-cancer immune response in rare cancers, we need to increase information by pooling big data from common cancers while utilizing small data from rare cancers. We propose to build a big-data warehouse containing molecular features and immunological parameters from patients with rare and common cancer and by applying machine learning algorithms to this large-scale warehouse, we can reliably identify molecular features associated with immunological parameters potentially relevant to the cancer-immune setpoint. These molecular features may contain targetable components that could ultimately lead to enhanced anti-cancer immune responses in patients with rare or common cancers. This step will also enable us to select molecular features with the highest likelihood of contributing to accurate predictive models for response on immunotherapy.

Brief summary of progress / results February 2023

We are investigating the factors that contribute to the cancer-immune setpoint in rare cancers with the goal of improving patient selection for immune checkpoint inhibitors and developing new treatment strategies. To do this, we have constructed a data warehouse containing transcriptomic profiles of tumor samples from both common and rare cancers. We are also developing machine learning tools to use these profiles to predict a patient's likelihood of responding to immunotherapy in rare cancers. So far, we have revealed that the transcriptional activity of metabolic processes is associated with drug sensitivities and immunological parameters in over 34,000 tumor samples, and have built a framework to predict functions for 55,000 coding and noncoding transcripts to identify genes involved in immunological processes and phenotypes using ~150,000 transcriptomes and > 23,000 gene sets.

Brief summary of progress / results January 2022

We try to identify the factors and how they interact to define the cancer-immune setpoint in rare cancers. This will improve patient selection for immune checkpoint inhibitors and help develop new treatment strategies to lower the cancer-immune setpoint, thereby enhancing anti-cancer immune responses on ICIs in rare cancers.

We constructed a data warehouse containing transcriptomic profiles of tumor samples of common and rare cancer.
We are developing several machine learning based tools that enables us to use these transcriptomic profiles to develop a predictive model that can be used to determine the likelihood of response on immunotherapy in rare cancers.

Finished results so far:

In >34,000 tumor samples, we revealed that transcriptional activity of metabolic processes was associated with drug sensitivities and immunological parameters. http://www. themetaboliclandscapeofcancer. com
Using ~150,000 transcriptomes and >23,000 gene sets, we built a framework based on independent components analysis in a guilt-by-association framework to predict functions for 55,000 coding and non- coding transcripts to identify genes involved in immunological processes and phenotypes. http://www. genetica-network.com

Brief summary of progress / results year 1

We try to identify factors within this project and how they interact to define the cancer-immune setpoint in rare cancers. This will improve patient selection for immune checkpoint inhibitors and help develop new treatment strategies to lower the cancer-immune setpoint, thereby enhancing anti-cancer immune responses on ICIs in rare cancers. For this, we make use of transcriptomic data obtained from public repositories. We started working on constructing a big-data warehouse containing ~500,000 transcriptomic profiles. The profiles are generated from samples obtained from patients with all kinds of disorders (e.g., patients with common and rare cancer and inflammatory diseases) and healthy individuals. We have and are developing tools based on this big-data warehouse that will enable us to gain insight into the cancer biology of rare-cancers, focusing on the anti-cancer immune setpoint.