A team of engineers and cancer researchers at Johns Hopkins University has developed a deep learning technique that can accurately predict cancer-associated protein fragments that can trigger an immune system response. If this technology succeeds in clinical trials, it could address significant challenges in the development of personalized immunotherapies and vaccines.
In a study published in the journal on July 20, nature machine intelligence, Johns Hopkins Biomedical Engineering, Johns Hopkins Institute for Computational Medicine, Johns Hopkins Kimmel Cancer Center, and Bloomberg Kimmel Institute for Cancer Immunotherapy, have found that a deep learning method called BigMHC can target cancer cells. We have shown that the above protein fragments can be identified. This is an essential step in understanding responses to immunotherapy and developing personalized cancer treatments.
“Cancer immunotherapy is designed to activate a patient’s immune system to destroy cancer cells,” says professor of biomedical engineering, oncology and computer science and center of the Institute for Computational Medicine. Member Dr. Rachel Curchin says: “A critical step in this process is the recognition of cancer cells by the immune system through the binding of T cells to cancer-specific protein fragments on the cell surface.”
The oncoprotein fragments that trigger this tumor-killing immune response may result from alterations (or mutations) in the genetic makeup of cancer cells called mutation-associated neoantigens. Each patient’s tumor has a unique set of such neoantigens that determine the tumor’s foreignness, in other words, how different the composition of the tumor is compared to self. By analyzing the cancer genome, scientists can identify which mutations carry the neoantigens associated with a patient’s tumor. By identifying those most likely to trigger a tumor-killing immune response, scientists can develop personalized cancer vaccines and customized immunotherapies, and inform patient choices for these therapies. It is possible. However, current methods for identifying and validating neoantigens that elicit immune responses are usually time-consuming and expensive, relying on laborious wet laboratory experiments.
Neoantigen validation is resource-intensive, so there is little data to train deep learning models. To address this, the researchers trained BigMHC, a set of deep neural networks, in his two-step process called transfer learning. First, BigMHC learned how to identify cell surface-presented antigens, an early stage of the adaptive immune response for which much data is available. BigMHC was then fine-tuned by learning T-cell recognition, a later stage for which few data exist. In this way, researchers leveraged the vast amount of data to build an antigen presentation model and refine this model to predict immunogenic antigens.
The researchers tested BigMHC on large independent datasets and showed that it is better at predicting antigen presentation than other methods. They further tested BigMHC based on data from study co-author Kelly Smith, Ph.D., associate professor of oncology at the Bloomberg Kimmel Institute for Cancer Immunotherapy, in identifying the neoantigens that BigMCH triggers T cells. We found that it significantly outperformed the other seven methods. response. “BigMHC has excellent accuracy in predicting immunogenic neoantigens,” he says.
“There is an urgent unmet clinical need to tailor immuno-oncology therapies to the subset of patients where they are most likely to respond, and BigMHC will identify the cancer hallmarks that drive tumor xenobiotics and demonstrate efficacy. It can provoke a significant anti-tumor immune response,” said the co-authors. -Author Valsamo “Elsa” Anagnostou, M.D., Ph.D., Director of Thoracic Oncology Biorepository, Johns Hopkins University Molecular Oncology Committee and Leader of Precision Oncology Analysis, Associate Professor of Oncology, Kimmel Cancer Center.
The research team is currently conducting several immunotherapy clinical trials to determine whether BigMHC can help scientists sift through hundreds of thousands of neoantigens and narrow them down to those that are most likely to elicit an immune response. We are expanding our efforts to test BigMHC.
“We hope that BigMHC will guide cancer immunologists to develop immunotherapies that can be used in multiple patients, as well as to develop personalized vaccines that boost a patient’s immune response to kill cancer cells. ,” says lead author Benjamin Alexander Albert. I was an undergraduate researcher in the Department of Biomedical Engineering and Computer Science at Johns Hopkins University when this research was conducted. Albert is currently pursuing his Ph.D. He is a student at the University of California, San Diego.
Karchin and her team believe that BigMHC and such machine-learning-based tools will provide the vast amount of data clinicians and cancer researchers need to develop more personalized approaches to cancer treatment. We believe it will help you sort efficiently and cost-effectively. “Deep learning will play an important role in cancer clinical research and practice,” says Curchin.
Reference: “Deep Neural Networks Predict Class I Major Histocompatibility Complex Epitope Presentation and Learn Neoepitope Immunogenicity” Benjamin Alexander Albert, Yunxiao Yang, Xiaoshan M. Shao, Dipika Singh, Kellie N Smith, Valsamo Anagnostou, Rachel Karchin, 20 July 2023, nature machine intelligence.
Co-authors of the study are Yunxiao Yang, Xiaoshan Shao and Dipika Singh of Johns Hopkins.
This work was partially supported by: National Institutes of Health (Grant CA121113), the Department of Defense Congressional Medical Research Program (Grant CA190755), and the ECOG-ACRIN Center for Integrated Translational Sciences on Thoracic Malignancies (Grant UG1CA233259).
Under a licensing agreement between Genentech and Johns Hopkins University, Mr. Xiao, Mr. Curchin, and the University will be entitled to a share of royalties related to the MHCnuggets neoantigen prediction technology. This arrangement has been reviewed and approved in accordance with the Johns Hopkins University Conflict of Interest Policy. Anagnostow has received research funding for her institution over the past five years from Bristol-Her Myers Squibb, AstraZeneca, Personal-Her Genome-Her Diagnostics, and Delphi-Her Diagnostics. . She is an advisory board member for Neogenomics and AstraZeneca. She is the inventor of several patent applications filed by Johns Hopkins University related to cancer genomic analysis, ctDNA therapeutic response monitoring, and immunogenomic characterization of response to immunotherapy, these being her 1 Licensed to one or more entities. Under the terms of these license agreements, the university and inventors are entitled to a share of fees and royalties.