A team at Johns Hopkins engineers and cancer researchers have developed deep-learning technology that predicts cancer-related protein fragments that can trigger an immune system response. If validated in clinical trials, the technology could help scientists overcome a major obstacle to developing personalized immunotherapies and vaccines.
A study published in the journal Nature Machine Intelligence cites researchers from Johns Hopkins Biomedical Engineering, the Johnd Hopkins Institute for Computational Medicine, and the BloombergKimmel Institute of Cancer Immunotherapy, who demonstrate that BigMHC, a deep-learning technique, can identify protein fragments on cancer cells that trigger invasive immune responses, which is crucial for understanding immunotherapy response and developing personalized cancer treatments.
The aim of cancer immunotherapy is to activate the immune system and destroy cancer cells. This process involves recognizing cancerous cells through T cell binding to specific protein fragments on the cell surface, which is essential for survival.
Changes in the genetic makeup of cancer cells can lead to the development of tumor-killing immune responses, which are triggered by mutations or changes in protein composition.
The limited resources available for neoantigen validation necessitated the use of deep neural networks for transfer learning, which led to the development of two-stage transfer training called BigMHC. This training involved identifying antigens at the cell surface, an early stage of the adaptive immune response, followed by fine-tuning T-cell recognition to refine a model of presentation that could predict immunogenic antiGENs.
The researchers examined BigMHC on a vast independent data set and found it to be more effective than other methods in predicting the appearance of antigens. They also tested it on data from study co-author Kellie Smith, Ph.D., an associate professor of oncology at the BloombergKimmel Institute for Cancer Immunotherapy, and discovered that BigMCH was significantly more successful than seven other techniques at identifying neoantigens that trigger T-cell response. Karchin emphasizes: “BigMACH used both have very precise precision when detecting
BigMHC can provide insight into cancer features that drive tumor foreignness and promote an effective anti-tumor immune response, as stated by Valsamo “Elsa” Anagnostou, a co-author of the study.
The team is extending their scope to include BigMHC in multiple immunotherapy clinical trials to assess its suitability for helping scientists filter through thousands of neoantigens and identify those that are most likely to trigger an immune response.
BigMHC is expected to assist cancer immunologists in creating immunotherapies that can be used for multiple patients or developing personalized vaccines, according to lead author Benjamin Alexander Albert. Albert is now a Ph.D. student at the University of California, San Diego.
Karchin and her team believe that Deep learning has a crucial role in clinical cancer research and practice by enabling clinicians and cancer researchers to efficiently and affordably sift through massive data sets.
Yunxiao Yang, Xiaoshan Shao, and Dipika Singh from Johns Hopkins were all co-authors in the study.
The work was aided in part by grants from the National Institutes of Health (CA121113), the Department of Defense Congressionally Directed Medical Research Programs (ca190755), and the ECOG-ACRIN Thoracic Malignancies Integrated Translational Science Center (grant UG1CA233259).
The Johns Hopkins University has approved a royalty distribution agreement with Genentech for MHCnuggets neoantigen prediction technology. This arrangement has been reviewed and approved by the university in accordance with its conflict-of-interest policies. Anagnostou is an advisory board member for Neogenomics and Astra Zeneca. She is also an inventor on patent applications related to cancer genomic analyses, ctDNA therapeutic response monitoring and immunogenomic features of response to immunotherapy that have been licensed to one or more of these agreements.