Jolanda de Vries & Li Xue

Radboudumc

In short, Jolanda de Vries's and Li Xue's research entails the following: 

Using geometric deep learning to identify MHC-II peptide vaccine candidates for CMMRD patients

In this project ‘Using geometric deep learning to identify MHC-II peptide vaccine candidates for CMMRD patients’, researchers of Radboudumc under guidance of Jolanda de Vries and Li Xue aim to develop a preventive cancer vaccine for patients with a paediatric tumour predisposition syndrome called Constitutional Mismatch Repair Deficiency (CMMRD).

CMMRD is a hereditary syndrome. Patients with CMMRD have a high risk of cancer due to mutations in both copies of DNA mismatch repair genes. These mutations cause cells to make altered proteins. These altered proteins can be recognized by the immune system.

We analyzed the DNA code of 42 tumours from 17 CMMRD patients. We compared this code with the DNA code of healthy cells from the same patient and detected more than 10,000 tumour-specific changes. In this Hanarth project, we use AI to predict which tumour-specific changes can trigger T cells to attack tumours. Once we know which altered proteins are eliciting an immune response against the tumour cells, we want to induce these responses by using dendritic cells. These cells play a central role in the immune system.

Our ultimate goal is to develop a preventive dendritic cell vaccine for CMMRD patients. Therefore, tumour-specific neoepitopes for the induction of anti-tumour responses are needed. Most cancer vaccine research focuses on CD8 T-cell responses, whereas CD4 T-cell responses are also required for effective responses. Here we will use innovative deep learning approaches to predict MHC class II-binding neoepitopes that are shared among CMMRD patients.

 

Publicaties

• Geometric deep learning improves generalizability of MHC-bound peptide predictions
  • Communications Biology 7, December 19, 2024
  • Click here to view the publication.
• Predicting reverse-bound peptide conformations in MHC Class II with PANDORA
  • Frontiers in Immunology, March 24, 2025
  • Click here to view the publication.

 

Short summary of progress/results 2025

Constitutional Mismatch Repair Deficiency (CMMRD) is a rare, severe cancer predisposition syndrome affecting children, caused by bi-allelic mutations in DNA mismatch repair genes. Current screening methods do not reliably detect cancers at an early, curable stage. Therefore, a preventive treatment strategy is urgently needed.

Our project focuses on developing a dendritic cell (DC)-based vaccine targeting tumor-specific frameshiftderived peptides (FSPs), which arise due to the  high mutation rates in CMMRD tumors. These FSPs represent highly specific, tumor-exclusive antigens and are ideal targets for immunotherapy. A key challenge is the prediction of FSP binding to MHC-II molecules, crucial for CD4 T-cell activation. 

We applied advanced deep learning models and updated 3D-structure prediction tools to improve MHC-II epitope prediction. Initial analyses of 44 tumor samples identified over 1,200 shared FSPs. We have begun validating FSP-specific CD4 T-cell responses in vitro, using both peptide libraries and epitope-specific dextramer staining.

Ultimately, this project aims to establish the feasibility of preventive DC vaccination in young, immunocompetent CMMRD patients, paving the way for early immunological interventions to prevent cancer development

 

Short summary of progress/results 2024

In this project ‘Using geometric deep learning to identify MHC-II peptide vaccine candidates for CMMRD patients’, researchers of Radboudumc under guidance of Jolanda de Vries and Li Xue aim to develop a preventive cancer vaccine for patients with a paediatric tumour predisposition syndrome called Constitutional Mismatch Repair Deficiency (CMMRD).

CMMRD is a hereditary syndrome. Patients with CMMRD have a high risk of cancer due to mutations in both copies of DNA mismatch repair genes. These mutations cause cells to make altered proteins. These altered proteins can be recognized by the immune system.

We analyzed the DNA code of 44 tumours from 19 CMMRD patients. We compared this code with the DNA code of healthy cells from the same patient and detected more than 10,000 tumour-specific changes. In this Hanarth project, we use AI to predict which tumour-specific changes can trigger T cells to attack tumours. Once we know which altered proteins are eliciting an immune response against the tumour cells, we want to induce these responses by using dendritic cells. These cells play a central role in the immune system.

Our ultimate goal is to develop a preventive dendritic cell vaccine for CMMRD patients. Therefore, tumourspecific neoepitopes for the induction of anti-tumour responses are needed. Most cancer vaccine research focuses on CD8 T-cell responses, whereas CD4 T-cell responses are also required for effective responses. Here we will use innovative deep learning approaches to predict MHC class II-binding neoepitopes that are shared among CMMRD patients.