Work

NeoAtlas-Tumor and NeoBert: A Database and A Predictive Model for Canonical and Noncanonical Tumor Neoantigens

Year: 2025

Type: article

Abstract: Neoantigens are classified into canonical and noncanonical types. Noncanonical neoantigens include those derived from noncoding regions, transposable elements (TE), and intron retention events, and they have recently gained considerable attention in cancer immunity. In this study, we focused on neoantigens presented by HLA class I molecules, which are central to CD8+ T cell-mediated immune responses. We curated 39,347 non-redundant neoantigen-HLA pairs from 14 immunopeptidomes studies, by analyzing unique features and differences across various sources of neoantigens. This knowledge enabled us to develop machine learning models for the prediction of different types of neoantigens. Our data and models are available at a public portal (https://ngdc.cncb.ac.cn/neoatlas) to facilitate broad access and future research. This resource offers advanced functionalities, including integration with epigenome browsers which allow easy navigation of epigenomic datasets to support and confirm the expression of neoantigens. We further demonstrate that combining our database with mass spectrometry analysis can identify noncanonical neoantigens. The resource we constructed holds significant value and promise for the development of neoantigen-based vaccines. All data, machine learning models, and analytical tools are freely available at the NeoAtlas-Tumor portal (https://ngdc.cncb.ac.cn/neoatlas).

Source: Genomics Proteomics & Bioinformatics

DOI: 10.1093/gpbjnl/qzaf105