The body's defense relies on the immune and lymphatic systems. The immune system acts as the body's personal police force, hunting down and eliminating pathogenic villains."The immune system of the body is very adept at identifying abnormal cells. Federica Eduati, a professor at TU/e's department of Biomedical Engineer, says that these cells could become cancerous or tumor-forming in the future. "Once cells are detected, the immune system attacks and kills them."Stopping the AttackIt's not always easy, as tumor cells may find ways to hide from the immune system.Tumor cells can also block the natural immune system. Oscar Lapuente Santana, PhD researcher in Computational Biology, says that proteins on tumor cells can shut down the immune system and put them into sleep mode.There is an effective way to stimulate the immune system and restore anti-tumor immunity. It's called immunotherapy.AdvertisementImmunotherapy:Immunotherapy, a treatment for cancer that aids the immune system to fight off cancer cells, is called immunotherapy. Immunotherapy can be done in two ways. One is immune checkpoint blocking drugs (ICB), which tell immune cells to ignore shutdown orders from cancer cells.The breakthrough of ICB is a revolutionary treatment for cancer. James P. Allison, Tasuku Honorjo were jointly awarded the 2018 Nobel Prize in Physiology or Medicine.Although ICB can be used to treat many patients with different types of cancer, only one third of patients respond to it.Eduati says that ICB has had a significant impact on patients, but could have a greater impact if we could quickly identify which patients are most likely respond to the treatment. "It would be wonderful if we could also understand why some patients don't respond to ICB."Lapuente-Santana, Eduati and their colleagues Maisa van Geenen (TU/e), Peter Hilberss (TU/e), and Francesca Finotello from the Medical University of Innsbruck used machine learning to solve this problem. Their research was published in Patterns.AdvertisementThe tumor microenvironmentResearchers first had to identify biomarkers in the tumor samples of patients to predict whether they would respond to ICB.Lapuente-Santana explains that tumors are more than just cancer cells. They also contain a variety of immune cells and fibrilli, which may have a pro-, anti-, or both, role and can interact with one another. We needed to understand how the complex regulatory mechanisms of the tumor microenvironment impact ICB response. To provide a high level representation of many aspects of the tumor microenvironment, we used RNA-sequencing data.The team used computational algorithms and data from clinical patient care to search for the biomarkers that could be used to predict the response of patients to ICB.Eduati says that although RNA-sequencing data are publically available, information about which patients received ICB therapy is limited to a subset of patients. We used a trick to solve this data problem.The secret is to get thereInstead of searching for the biological response to ICB treatment itself, the researchers selected several alternative immune responses from the same data sets. Despite not being the primary reaction to ICB treatment, they can be combined to indicate the effectiveness of ICB.This approach allowed the team to use large public data sets with thousands of patient samples in order to train robust machine learning models."The proper training of machine learning models was a significant challenge in this project. "We were able solve this problem by looking at substitute immune reactions during the training process," says Lapuente Santana.The researchers had the machine learning models ready and tested its accuracy on different datasets that contained the actual response to ICB treatments. Eduati says, "We found that our machine learning model outperforms current biomarkers used in clinical settings for assessing ICB treatment."Why are Lapuente-Santana and Eduati turning to mathematical models for medical treatment? This will replace doctors? "Mathematical models are able to give a large picture of the interconnectedness of individual cells and molecules, as well as approximate the behavior of cancerous cells in a patient. This allows for personalized immunotherapy treatments with ICB in clinical settings. The models may help doctors make the right decisions about the best treatment but they will not replace them. Eduati.The model can also help to identify the important biological mechanisms that mediate the biological response. The model can help you decide how to best combine ICB and other treatments to increase its clinical effectiveness. Before these findings can be translated into clinical settings, they will need to be validated experimentally.Dare to DreamSome of the researchers used the machine learning approach described in the paper to participate in a DREAM contest called "Anti-PD1 Response Prediction DREAM Contest."DREAM is a crowd-sourced organization that runs challenges involving algorithms and biomedicine. Eduati adds that "we came first in one sub-challenge and competed under cSysImmunoOncoteam."Although our immune system is a skilled detective and disease hunter it sometimes needs help to eliminate elusive villains such as cancer cells. Although immunotherapy with immune checkpoint blocking agents is one option, it may not work for everyone.Lapuente-Santana and Eduati have definitely dreamed, and their efforts could prove crucial in identifying future ICB patients.The researchers are able to quickly deliver effective and appropriate cancer treatments to patients by using machine learning.For some cancer cells, this means there may not be a safe place to hide or run.
