Drastic improvement in MHC class I epitope prediction through training data augmentation, generative adversarial network modeling of selfness, and consideration of homology to class II epitopes.

Accurate identification of immunogenic neoantigens is critical for personalized cancer vaccines but remains hindered by the high false-positive rates of current predictors. Here, we present DeepOmicsNeo, a modular deep learning framework that harmonizes three biological determinants of immunogenicity. The first module, DeepOmicsNeo-Binding, overcomes data scarcity in non-9mer peptide-Class I Major Histocompatibility Complex (MHC) binding prediction via a novel augmentation strategy that synthesizes pseudo-binders and non-binders by editing non-anchor residues, outperforming its competitors. The second module, DeepOmicsNeo-Selfness, addresses the paucity of negative training data for immunogenicity prediction by utilizing a Generative Adversarial Network trained on around 60 million human self-peptides. This unsupervised approach quantifies selfness to detect immunogenic peptides and the resulting score correlated with IFN-γ secretion by T-cells. The third module, DeepOmicsNeo-H2, exploits the central-region homology between known HLA Class-I and Class-II restricted epitopes. The integrated model, DeepOmicsNeo-Immunogenicity, outperforms state-of-the-art predictors on the TESLA, HiTIDE, and TiTANIAN benchmark datasets.[세오2.1] Top20 precision, the most important benchmark for clinical utility in personalized cancer vaccine design, was drastically improved over other predictors. Together, these findings indicate that jointly modeling the three complementary features captures biologically meaningful determinants of immunogenicity, supporting DeepOmicsNeo as a robust computational framework for neoantigen prioritization in personalized cancer immunotherapy.

Authors

Institutions

Publication Details

Journal
bioRxiv (Cold Spring Harbor Laboratory)
Published
2026-11-18
DOI
https://doi.org/10.64898/2026.11.16.766890
Primary Topic
vaccines and immunoinformatics approaches
Type
preprint

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Drastic improvement in MHC class I epitope prediction through training data augmentation, generative adversarial network modeling of selfness, and consideration of homology to class II epitopes.

Hae-Suk Kim, Sejin Oh, Soonmyung Paik, Seong-Eui Hong
1 links
bioRxiv (Cold Spring Harbor Laboratory)
vaccines and immunoinformatics approaches
preprint

Drastic improvement in MHC class I epitope prediction through training data augmentation, generative adversarial network modeling of selfness, and consideration of homology to class II epitopes.

Hae-Suk Kim, Sejin Oh, Soonmyung Paik, Seong-Eui Hong
preprint en

Abstract

Accurate identification of immunogenic neoantigens is critical for personalized cancer vaccines but remains hindered by the high false-positive rates of current predictors. Here, we present DeepOmicsNeo, a modular deep learning framework that harmonizes three biological determinants of immunogenicity. The first module, DeepOmicsNeo-Binding, overcomes data scarcity in non-9mer peptide-Class I Major Histocompatibility Complex (MHC) binding prediction via a novel augmentation strategy that synthesizes pseudo-binders and non-binders by editing non-anchor residues, outperforming its competitors. The second module, DeepOmicsNeo-Selfness, addresses the paucity of negative training data for immunogenicity prediction by utilizing a Generative Adversarial Network trained on around 60 million human self-peptides. This unsupervised approach quantifies selfness to detect immunogenic peptides and the resulting score correlated with IFN-γ secretion by T-cells. The third module, DeepOmicsNeo-H2, exploits the central-region homology between known HLA Class-I and Class-II restricted epitopes. The integrated model, DeepOmicsNeo-Immunogenicity, outperforms state-of-the-art predictors on the TESLA, HiTIDE, and TiTANIAN benchmark datasets.[세오2.1] Top20 precision, the most important benchmark for clinical utility in personalized cancer vaccine design, was drastically improved over other predictors. Together, these findings indicate that jointly modeling the three complementary features captures biologically meaningful determinants of immunogenicity, supporting DeepOmicsNeo as a robust computational framework for neoantigen prioritization in personalized cancer immunotherapy.

bioRxiv (Cold Spring Harbor Laboratory)
Tech University of Korea (KR), Theragen (KR)
Korea Health Industry Development Institute, Advanced Research Projects Agency
Good health and well-being
vaccines and immunoinformatics approaches
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon