Summary: Johns Hopkins researchers applied artificial intelligence to build a comparative map of T-cell surface receptors, revealing patterns that may predict and inform responses to immunotherapy and vaccine strategies.
Source: Johns Hopkins Medicine.
Johns Hopkins scientists have developed an AI-driven mapping tool, called ImmunoMap, to compare and visualize types of T-cell receptors—the chemical “antennas” on immune T-cells that recognize antigens. Laboratory studies using mouse and human T-cells suggest people with cancer who possess a broader variety of these receptors may be more likely to respond to immunotherapy drugs and vaccines.
The Methods and early results were reported Dec. 20 in Cancer Immunology Research. ImmunoMap combines high-performance computing, sequence analysis, and unsupervised machine learning to turn raw T-cell receptor (TCR) sequencing data into biologically meaningful maps of receptor relatedness and potential functional specificity.
“ImmunoMap gives scientists a picture of the wide diversity of the immune system’s responses to cellular antigens,” says Jonathan Schneck, M.D., Ph.D., professor of pathology, medicine and oncology at Johns Hopkins University School of Medicine and a member of the Kimmel Cancer Center.
T-cell receptors detect antigens—fragments of proteins that can indicate infection, malignancy, or other changes in the body. When TCRs recognize a foreign or abnormal antigen, they trigger immune activation and recruit other immune cells. Tumors often evade immune detection by altering or hiding antigens, so understanding patterns of TCR recognition is central to designing targeted immunotherapies that spare healthy tissue.
“Much of immunotherapy today is built on the premise that we know these antigens,” says Johns Hopkins biomedical engineering M.D./Ph.D. student John-William Sidhom. “But we actually don’t know as much as we need to about them and the T-cells that recognize them.”
To address this gap, Sidhom and colleagues created a mathematical model that converts receptor sequence data into numeric distances reflecting sequence similarity. Using an unsupervised learning algorithm, ImmunoMap assigns short distance ranks between similar receptor sequences and longer ranks between more divergent sequences. These distance metrics let the AI cluster receptors that are likely to share antigen specificity, producing a visual and quantitative repertoire map.
After transforming thousands of sequences into distance-based metrics, the system detects patterns across the repertoire and highlights groups of related receptors. “T-cell receptors that are very similar, with slight differences in their sequences, may be recognizing the same antigen,” Schneck explains. ImmunoMap therefore captures both sequence relatedness and potential shared function, giving researchers a clearer view of immune diversity.
The team tested ImmunoMap on clinical TCR sequencing data from tumor-infiltrating lymphocytes (TILs) collected from 34 cancer patients enrolled in a nationwide trial of the anti–PD-1 immunotherapy nivolumab. Among 34 patients with melanoma, three responded to nivolumab while the remainder did not. Responders displayed more receptor clusters—an average of 15—compared with eight to nine clusters in non-responders. The researchers also observed a 10–15 percent reduction in receptor diversity among responders four weeks after treatment, consistent with expansion of specific T-cell clones that target tumor antigens.
“Those patients had a broad array of receptor weaponry before their treatment, which may have allowed the right receptor to kill their cancer cells,” Schneck says. As effective clones expanded, overall structural diversity fell—an expected consequence when particular T-cells proliferate after recognizing tumor antigens.
While many researchers emphasize the importance of T-cell infiltration into tumors, the Johns Hopkins group stresses that infiltration alone does not fully explain variable responses to immunotherapy. ImmunoMap suggests repertoire diversity and the presence of appropriate receptor clusters also play a critical role.
The researchers also mapped TCR diversity in mouse models, comparing T-cells specific for tumor-associated antigens in mice with and without tumors. Consistently, receptor diversity decreased in samples collected closer to the tumor site, suggesting localized selection or expansion of specific clones and offering clues about tumor immune evasion mechanisms.
Schneck notes that additional ImmunoMap data are necessary before the tool can reliably predict individual patient responses to immunotherapy. “At this point, ImmunoMap can’t match T-cell receptors to specific antigens, or determine whether those antigens are important for immunotherapy response in any individual patient,” he says. Nonetheless, the team hopes ImmunoMap will eventually help design better vaccines and engineered T-cell therapies by identifying receptor patterns linked to effective antitumor immunity.
Contributors to the research include John-William Sidhom, Catherine A. Bessell, Alyssa Kosmides, Jonathan J. Havel, Timothy A. Chan and Jonathan P. Schneck.
Funding: Support came from the Johns Hopkins University–Coulter Translational Partnership, the TEDCO Maryland Innovation Initiative, the Troper Wojcicki Foundation, the National Institutes of Health (grants R01-AI44129, CA108835 and U01-AI113315), Bristol Myers-Squibb, the Pershing Square Sohn Cancer Research Alliance, the PaineWebber Chair, Stand Up To Cancer and the Starr Cancer Consortium.
Conflict of interest: Jonathan Schneck has a licensing agreement with NexImmune and may receive royalties through Johns Hopkins University; he served as a founder and is a scientific advisory board member of NexImmune and holds equity. Timothy A. Chan is a cofounder of Gritstone Oncology. Jonathan Havel’s spouse is employed full-time by Regeneron Pharmaceuticals. These relationships were reviewed and approved by Johns Hopkins University in accordance with its conflict-of-interest policies.
ImmunoMap: A Bioinformatics Tool for T-Cell Repertoire Analysis
ImmunoMap applies a phylogenetics-inspired sequence analysis to TCR repertoires to reveal relatedness and functional clustering. In preclinical models, ImmunoMap distinguished repertoire differences between self- and foreign-antigen responses and detected organ-specific TCR patterns in tumor-bearing mice. In clinical TIL data from patients treated with anti–PD-1, ImmunoMap—unlike standard sequence metrics—identified predictive signatures in pre- and post-therapy samples that correlated with therapeutic response. The tool offers a new approach to interpret growing TCR sequencing datasets and may guide future immunotherapy and vaccine design.