imNEO: High-accuracy neoantigen prediction through AI-based integrated analysis of multiple immunogenicity-related factors.

Abstract

e14639 Background: Neoantigens have emerged as compelling targets for cancer vaccines, with early clinical trials confirming their safety and efficacy. Accurate prediction of neoantigens is critical for the successful development of personalized cancer vaccines. Here, we present imNEO, a highly accurate neoantigen prediction platform that integrates multiple immunogenicity-related factors. Methods: Experimentally validated human tumor neoantigens were collected. Previously reported immunogenicity-related factors, along with newly developed factors, were tested for their association with immunogenicity. Integrated machine learning models were constructed and evaluated using data from melanoma, lung cancer, and gastrointestinal cancers. To verify prediction accuracy, immunogenicity using IFN-γ ELISPOT assays, tumor growth inhibition, and antibody secretion using ELISA were assessed in a colon cancer mouse model. Results: A total of 30 factors, encompassing epitope properties, antigen processing and presentation, T-cell interaction, the tumor microenvironment, and the differential index between mutant and wild-type, were identified as being associated with the immunogenicity of neoantigens. Seven machine learning algorithms were integrated to construct the immunogenicity prediction model. Across multiple human cancer datasets, imNEO demonstrated up to a 15.6% increase in the area under the precision-recall curve compared to existing methods. Notably, positive predictive values for the top 10 and 20 ranked neoantigens were 85% and 70% on average, respectively. In immunogenicity tests, 19 out of 25 (76%) neoantigens identified from the MC38 colon cancer model induced strong IFN-γ secretion, confirming a high immunogenicity rate suitable for application in personalized cancer vaccines. In tumor growth inhibition and antibody secretion tests, neoantigen-treated groups exhibited significantly reduced tumor growth (p = 1.26E-05) and increased IgG antibody secretion, demonstrating that imNEO-predicted neoantigens possess the necessary characteristics for effective tumor suppression. Conclusions: imNEO significantly advances neoantigen prediction, integrating immunogenicity-related factors to achieve superior accuracy. Experimental validation confirmed high immunogenicity and tumor-suppressive effects, highlighting its potential for developing effective personalized cancer vaccines.