Abstract
BACKGROUND: Tumor immunogenicity is a concept for modeling the susceptibility of tumors to immune checkpoint inhibitors (ICIs) and other immunotherapies. Single biomarkers, such as tumor mutation burden (TMB) or PDL1 expression, have been shown to correlate with ICI outcomes but are poor predictors of overall and progression-free survival (OS, PFS). Complex machine learning models that integrate multiple biomarkers have shown improved predictions but often lack clear a priori interpretability. In this study, we developed a coherent Multi-Omics Tumor Immunogenicity score (MOTIscore) that combines immunogenicity biomarkers derived from genomic and transcriptomic data and demonstrated its generalizability across multiple cancer types. METHODS: Several immunogenicity biomarkers, including TMB, neoantigen burden, T-cell receptor repertoire, PDL1 expression, B2M expression, and variants in pathways of ICI response and resistance, were integrated using a weighted sum scoring scheme. The weights were determined using statistical tests in a large melanoma ICI cohort. We compared the MOTIscore with a machine learning (ML) model trained using the same biomarkers and evaluated the model using melanoma, gastric cancer, and pan-cancer datasets. RESULTS: MOTIscore achieved results similar to those of the ML model in predicting ICI in melanoma and gastric cancer, with both outperforming TMB. Gastric cancer and melanoma patients with high MOTIscores had a significantly extended overall and progression-free survival. Gene set enrichment analysis revealed the enrichment of immune-related pathways in patients with high MOTIscores. Differential expression analysis between patients with high and low immunogenicity identified highly expressed C-X-C motif chemokine ligands as important characteristics associated with successful ICI therapy and significantly improved PFS. MOTIscores varied widely across cancers treated in the molecular tumor board at our hospital and showed distinct distributions between non-immunogenic and immunogenic cancer types. CONCLUSIONS: MOTIscore demonstrated improved ICI outcome predictions compared to single-omics biomarkers. Patients with higher tumor immunogenicity also show significantly improved OS and PFS in melanoma and gastric cancer. The results demonstrate the potential use of the MOTIscore to prioritize ICI in personalized cancer treatment. However, ICI outcomes and survival should be investigated in prospective studies, and additional cancer types and larger patient cohorts are needed.