Diego Amaya-Ramirez

Personalized neoantigen-based vaccines have progressed from concept to clinical evidence for some of the most aggressive human tumors, yet a key challenge remains: determining which candidates should be prioritized for vaccine inclusion. Public neoantigens, shared across tumor subtypes and patient populations, offer potential as off-the-shelf candidates for broad-spectrum immunotherapy. Candidate selection relies on in silico pipelines that estimate binding; however, binding affinity alone is an imperfect marker of immunogenicity. Molecular dynamics (MD) combined with endpoint free-energy methods estimates binding free energy directly from simulated trajectories, and hydrogen mass repartitioning (HMR) mitigates the computational cost of hydrogen-involving bond vibrations. This study assessed whether HMR-accelerated MD can prioritize immunogenic neoantigens through peptide–HLA binding free-energy calculations and quantified the trade-off between accuracy and computational cost. Ten immunogenic public neoantigens (seven paired with their wild-type counterparts and three frameshift neoantigens analyzed as unpaired references) were modeled in complex with their HLA-I alleles, built de novo using AlphaFold3 and propagated with a 4-fs HMR time step. Mean binding free energy was computed across ten replicas per complex, and each SNV pair was scored concordant with experimental data when the neoantigen formed a more favorable complex than its wild type. Free energies were recomputed at 100–500 ns to identify the shortest trajectory preserving reliable ranking. Free-energy calculations reproduced experimental concordance for six of seven SNV pairs and were inconclusive for TP53 R8H; frameshift neoantigens yielded comparable values. Most complexes retained both peptide termini anchored to the HLA groove throughout, and RMSD and RMSF metrics agreed with one another and with the free-energy rankings. The PIK3CA E545K (E11K) pair showed the most distinctive signature: the peptides engaged the groove almost identically over P1–P8, with the entire difference confined to the C-terminal pocket, where loss of the wild-type’s anchor also disrupted its backbone anchor. This pair showed the largest free-energy separation in the panel, and restricting analysis to bound frames confirmed a substantial electrostatic and van der Waals advantage favoring the neoantigen. These findings support HMR-accelerated binding free-energy MD as a practical refinement step for neoantigen prioritization and its application to assessing public neoantigens for cancer vaccine design.

Approche science des données pour l'exploration de l'antigénicité HLA fondée sur les structures 3D et la dynamique moléculaire: Cette thèse présente une approche de science des données pour explorer l'antigénicité des molécules du complexe majeur d'histocompatibilité (HLA) en se basant sur leurs structures 3D et des simulations de dynamique moléculaire (MD). L'objectif principal est de mieux comprendre les déterminants du rejet de greffe causé par les anticorps spécifiques du donneur (DSA en anglais) et d'améliorer l'évaluation de la compatibilité donneur-receveur, qui ne prend actuellement pas en compte la structure 3D des antigènes HLA. La thèse a produit et analysé les structures 3D et les simulations MD de 207 antigènes HLA. Diverses méthodes de science des données ont été employées pour transformer les données 3D brutes en descripteurs statiques et dynamiques, permettant finalement l'entraînement d'un prédicteur d'épitopes HLA hautement performant. La thèse explore les propriétés structurales des antigènes HLA en utilisant selon divers points de vue, examinant des caractéristiques telles que l'accessibilité au solvant, les propriétés physico-chimiques et la flexibilité des chaînes latérales des acides aminés (AA). L'étude de la flexibilité des AA pendant les simulations MD a révélé que la flexibilité était plus faible pour les AA présents dans les éplets confirmés par rapport aux non-éplets, ce qui est cohérent avec des résultats récents sur les épitopes en général. De plus, la thèse introduit l'utilisation des polynômes de Zernike pour représenter la surface 3D des molécules, facilitant des tâches telles que la comparaison et la classification des structures. L'hypothèse selon laquelle l'antigénicité des HLA dépend de la similarité structurelle entre les antigènes HLA du receveur et du donneur a été étudiée en utilisant la distance de Wasserstein pour comparer des nuages de points multidimensionnels représentant les trajectoires MD. L'utilisation des polynômes de Zernike pour l'identification des conformères de peptides le long de trajectoires MD a également été explorée. Les résultats ont montré une capacité prometteuse à identifier des conformères distinctifs en fonction de leur forme. Une contribution clé de la thèse est la construction d'un prédicteur d'épitopes HLA appelé HLA-EpiCheck. En utilisant des caractéristiques structurales statiques et dynamiques extraites des simulations MD, un jeu de données composé de 18 descripteurs calculés sur près de 7000 patchs de surface 3D dérivés de 207 antigènes HLA a été construit. Le prédicteur a été entraîné en utilisant des éplets confirmés de la base de données HLA Eplet Registry, puis employé pour prédire le statut des éplets non confirmés. Les résultats de prédiction ont montré une cohérence remarquable avec des résultats expérimentaux. De plus, une analyse de l'importance des descripteurs, effectuée sur HLA-EpiCheck, a révélé l’importance notable des descripteurs associés à la flexibilité des chaînes latérales, soulignant l’intérêt d’introduire des descripteurs dynamiques dans le processus d’apprentissage quand il s’agit de structures de protéines. En outre, la thèse a impliqué le développement d'une base de données pour les structures 3D des antigènes HLA et d'une interface graphique appelée HLA-3D-Diff. Cette interface permet aux utilisateurs de superposer des structures 3D d'antigènes HLA, de visualiser leurs trajectoires dans les simulations de DM et d'identifier les différences potentielles apparaissant de manière dynamique pendant les trajectoires. Elle sert d'outil pour explorer manuellement les paires d'antigènes HLA dans des situations cliniques d'immunisation complexes ou inattendues. Le travail inclus dans cette thèse vise à permettre la prise en compte de caractéristiques dynamiques dans l’étude des différences structurales et des prédictions d'épitopes HLA lors de l'appariement donneur-receveur pour l'attribution des greffons.Les travaux futurs pourraient impliquer des études cliniques rétrospectives de la contribution de ces nouveaux éléments au maintien ou au rejet de la greffe

The target of an anti‐HLA antibody is an epitope on the surface of the antigen, and in particular the polymorphic eplet of its core, which contains one or a few polymorphic residues accessible to the molecule surface. The HLA Eplet Registry database references more than 550 HLA eplets thus deduced from AA sequence alignments. However, not all eplets have yet been verified and this list is not exhaustive. We performed a systematic analysis of sequence alignment of DQ antigens, to identify polymorphic amino acids so far not proposed as candidate eplets. From this, we describe and validate 3 DQB1 eplets targeted by sera of organ‐transplanted patients and explore a new eplet overlapping the DQA1*03 and DQB1*03 chains. Serum antibody profiles using LABScreen, Lifecodes and a complementary DQ Luminex single antigen bead panel all showed a concordant pattern incriminating these residues. Moreover, antibodies were adsorbed and eluted using human splenic mononuclear cells and HLA‐DQ transfected murine cell clones, thereby validating these four eplets. Their localisation by 3D modelling revealed uncertain accessibility of some residues implicated. The protrusion and accessibility of an eplet on the HLA surface may ultimately not always be sufficient to predict antibody binding, since the dynamic flexibility of the molecule can modify these parameters. Our strategy, combining Luminex single antigen assays, cell adsorption/elution and in silico prediction of surface exposure, altogether concur to ascertaining the realness of candidate eplets in the highly complex HLA system.

Motivation: The human leukocyte antigen (HLA) system is the main cause of organ transplant loss through the recognition of HLAs present on the graft by donor-specific antibodies raised by the recipient. It is therefore of key importance to identify all potentially immunogenic B-cell epitopes on HLAs in order to refine organ allocation. Such HLAs epitopes are currently characterized by the presence of polymorphic residues called “eplets”. However, many polymorphic positions in HLAs sequences are not yet experimentally confirmed as eplets associated with a HLA epitope. Moreover, structural studies of these epitopes only consider 3D static structures. Results: We present here a machine-learning approach for predicting HLA epitopes, based on 3D-surface patches and molecular dynamics simulations. A collection of 3D-surface patches labeled as Epitope (2117) or Nonepitope (4769) according to Human Leukocyte Antigen Eplet Registry information was derived from 207 HLAs (61 solved and 146 predicted structures). Descriptors derived from static and dynamic patch properties were computed and three tree-based models were trained on a reduced non-redundant dataset. HLA-Epicheck is the prediction system formed by the three models. It leverages dynamic descriptors of 3D-surface patches for more than half of its prediction performance. Epitope predictions on unconfirmed eplets (absent from the initial dataset) are compared with experimental results and notable consistency is found. Availability and implementation: Structural data and MD trajectories are deposited as open data under doi: 10.57745/GXZHH8. In-house scripts and machine-learning models for HLA-EpiCheck are available from https://gitlab.inria.fr/capsid.public_codes/hla-epicheck.

1 The Human Leukocyte Antigen (HLA) system is the main cause of organ transplant loss through the recognition of HLA proteins by Donor-Specific Antibodies (DSA). Therefore, the identification of potentially immunogenic epitopes is a key task to refine organ allocation and then to improve the survival of transplanted organs. Here, we present HLA-EpiCheck, a machine learning predictor for B-cell epitopes on HLA proteins that leverages an unprecedented dataset of high-quality molecular dynamics simulations of 207 HLA proteins. Candidate epitopes are represented as surface patches centered on solvent-accessible residues and described by a set of 18 descriptors. The descriptors include both static and dynamic properties, such as hydrophobicity, electrostatic charges, relative solvent-accessible surface area and side-chain flexibility. The HLA-EpiCheck was trained using an Extra Trees ensemble learning method and was compared to DiscoTope-3.0, a state-of-the-art B-cell epitope predictor. HLA-EpiCheck largely outperformed DiscoTope-3.0 in the task of predicting HLA epitopes. HLA-EpiCheck was also used to assess the epitope status of a subset of non-confirmed eplets. The predictions were compared to experimental data and a notable consistency was found. These results suggest that HLA-EpiCheck could be used to better define HLA matching between donor and recipient to reduce de novo DSA formation and graft rejection.

Personalized cancer vaccines based on neoantigens are a new and promising treatment for cancer; however, there are still multiple unresolved challenges to using this type of immunotherapy. Among these, the effective identification of immunogenic neoantigens stands out, since the in silico tools used generate a significant portion of false positives. Inclusion of molecular simulation techniques can refine the results these tools produce. In this work, we explored docking and molecular dynamics to study the association between the stability of peptide–HLA complexes and their immunogenicity, using as a proof of concept two HLA-A2-restricted neoantigens that were already evaluated in vitro. The results obtained were in accordance with the in vitro immunogenicity, since the immunogenic neoantigen ASTN1 remained bound at both ends to the HLA-A2 molecule. Additionally, molecular dynamic simulation suggests that position 1 of the peptide has a more relevant role in stabilizing the N-terminus than previously proposed. Likewise, the mutations may have a “delocalized” effect on the peptide–HLA interaction, which means that the mutated amino acid influences the intensity of the interactions of distant amino acids of the peptide with the HLA. These findings allow us to propose the inclusion of molecular simulation techniques to improve the identification of neoantigens for cancer vaccines.

Personalized cancer vaccines are presented as a new and promising treatment alternative for cancer, especially in those cases where effective treatments do not yet exist. However, multiple challenges remain to be resolved so that this type of immunotherapy can be used in the clinical setting. Among these, the effective identification of immunogenic peptides stands out, since the in-silico tools currently used generate a significant portion of false positives. This is where molecular simulation techniques can play an important role when it comes to refining the results produced by these tools. In the present work, we explore the use of molecular simulation techniques such as docking and molecular dynamics to study the relationship between stability of peptide-HLA complexes and their immunogenicity using two HLA-A2-restricted neoantigens that have already been evaluated in vitro . The results obtained agreed with the in vitro immunogenicity of the immunogenic neoantigen ASTN1 the only one that remains bound at both ends to the HLA-A2 molecule. Additionally, molecular dynamics indicates that position 1 of the peptide has a more important role in stabilizing the N-terminal part than previously assumed. Likewise, the results suggest that the mutations may have a “delocalized” effect on the peptide-HLA interaction, that is, they may modulate the intensity of the interactions of other amino acids in the peptide. These results highlight the suitability of this type of in silico strategy to identify peptides that form stable complexes with HLA proteins that are highly immunogenic for CD8+ T cells.

Introducción. Los avances que se han hecho en inmunoterapia contra el cáncer y la respuesta clínica de los pacientes que han recibido este tipo de terapia la han convertido en el cuarto pilar para el tratamiento del cáncer. Objetivo. Describir brevemente el fundamento biológico de la inmunoterapia personalizada contra el cáncer basada en neoantígenos, las perspectivas actuales de su desarrollo y algunos resultados clínicos de esta terapia. Materiales y métodos. Se realizó una búsqueda de la literatura en PubMed, Scopus y EBSCO utilizando la siguiente estrategia de búsqueda: tipo de artículos: estudios experimentales originales, ensayos clínicos y revisiones narrativas y sistemáticas sobre métodos de identificación de mutaciones generadas en los tumores y estrategias de inmunoterapia del cáncer con vacunas basadas en neoantígenos; población de estudio: humanos y modelos animales; periodo de publicación: enero 1989- diciembre 2019; idioma: inglés y español; términos de búsqueda: “Immunotherapy”, “Neoplasms”, “Mutation” y “Cancer Vaccines”. Resultados. La búsqueda inicial arrojó 1344 registros; luego de remover duplicados (n=176), 780 fueron excluidos luego de leer su resumen y título, y se evaluó el texto completo de 338 para verificar cuáles cumplían con los criterios de inclusión, seleccionándose finalmente 73 estudios para análisis completo. Todos los artículos recuperados se publicaron en inglés, y fueron realizados principalmente en EE. UU. (43.83%) y Alemania (23.65%). En el caso de los estudios originales (n=43), 20 se realizaron únicamente en humanos, 9 solo en animales, 2 en ambos modelos, y 12 usaron metodología in silico. Conclusión. La inmunoterapia personalizada contra el cáncer con vacunas basadas en neoantígenos tumorales se está convirtiendo de forma contundente en una nueva alternativa para tratar el cáncer. Sin embargo, para lograr su implementación adecuada, es necesario usarla en combinación con tratamientos convencionales, generar más conocimiento que contribuya a aclarar la inmunobiología del cáncer, y reducir los costos asociados con su producción. Introduction: The progress that has been made in cancer immunotherapy and the clinical response of patients who have undergone this type of therapy have made it the fourth pillar of cancer treatment. Objective: To briefly describe the biological rationale of personalized neoantigen-based cancer immunotherapy, the current perspectives regarding its development and some of the clinical outcomes obtained with this therapy. Materials and methods: A literature search was performed in PubMed, Scopus and EBSCO using the following search strategy: type of articles: original experimental studies, clinical trials, and narrative and systematic reviews addressing methods to identify mutations caused by tumors and cancer immunotherapy strategies based on neoantigen-based vaccines; study population: humans and animal models; publication period: January 1989 - December 2019; language: English and Spanish; search terms: “Immunotherapy”, “Neoplasms”, “Mutation” and “Cancer Vaccines”. Results: The initial search yielded 1344 records; after removing duplicates (n=176), 780 studies were excluded after reading their abstract and title, and the full text of 338 read to confirm which met the inclusion criteria, finally including 73 studies for full analysis. All articles retrieved were published in English, and were mainly conducted in USA (43.83%) and Germany (23.65%). In the case of original studies (n=43), 20 were performed in humans only, 9 in animals only, 2 in both models, and 12 used in silico methodology. Conclusion: Personalized cancer immunotherapy with tumor neoantigen-based vaccines is strongly becoming a new alternative to treat cancer. However, to achieve its appropriate implementation, it is necessary to use it in combination with conventional treatments, to produce more knowledge that helps clarify cancer immunobiology, and to reduce the costs associated with its production.