Abstract Precise neoepitope discovery is crucial for effective cancer therapeutic vaccines. Conventional approaches struggle to build a repertoire with sufficient immunogenic epitopes. We developed a workflow leveraging full-length ribosome–nascent chain complex–bound mRNA sequencing (FL-RNC seq) and artificial intelligence–based predictive models to accurately identify the neoepitope landscape, especially large-scale transcript variants (LSTVs) missed by short-read sequencing. In the MC38 mouse model, we identified 22 LSTV-derived neoepitopes encoded by a synthesized mRNA lipid nanoparticle vaccine. As a standalone therapy and combined with anti–PD-1 immunotherapy, the vaccine curbed tumor progression, induced robust T cell–specific immunity, and modulated the tumor microenvironment. This underscores the multifaceted potentials of LSTV-derived vaccines. Our approach expands the neoepitope source repertoire, offering a method for discovering personalized cancer vaccines applicable to a broader tumor range. The results highlight the importance of comprehensive neoepitope identification and the promise of LSTV-based vaccines for cancer immunotherapy. __________________________________________________________________ The success of cancer therapeutic vaccines hinges on precise and efficient discovery of tumor neoepitopes. INTRODUCTION Personalized cancer therapeutic vaccines have yielded encouraging results in several ongoing clinical trials and may represent a promising strategy for cancer immunotherapy ([54]1–[55]3). Neoepitopes, immunogenic parts recognized by T cell receptors (TCRs) on neoantigens formed by tumor-specific mutations, are pivotal for developing cancer vaccines. Their specificity stems from their tumor-exclusive presence, reducing the likelihood of an immune response against normal tissues and therefore could overcome the conventional obstacles of immune tolerance. This unique feature makes neoepitopes the most “favorable” immunogenic targets of T cell ([56]4). Personalized cancer vaccines based on neoepitopes have demonstrated good safety and efficacy profile. Neoepitopes act not only as potential therapeutic targets but also as useful biomarkers to monitor and predict the efficacy of immunotherapies. Clinical studies in recent years have revealed the importance of precise neoepitope identification in eliciting a strong tumor-specific T cell response in patients with favorable therapeutic outcomes, including partial response and even complete response ([57]5, [58]6). However, the neoepitope identification methods in these clinical studies were primarily tailored for cancers with high tumor mutation burden (TMB), such as melanoma and pancreatic ductal adenocarcinoma (PDAC) ([59]3, [60]7). Such methods are obviously not ideal for cancers with low TMB. For instance, neoepitope vaccines against glioblastoma have yielded suboptimal results with low response rates ([61]8). Traditional short-read sequencing techniques are considered unable to identify sufficient neoepitopes in low TMB tumors ([62]9), which largely limit effective vaccine development across multiple cancer indications. Therefore, establishing more powerful neoepitope discovery method is imperative for broadening the application of cancer vaccines to cancers with low TMB. An intricate mosaic of genomic aberrations gives rise to neoantigens, of which immunogenic parts presented by major histocompatibility complex (MHC) and recognized by TCRs are neoepitopes. Previous studies mainly focused on somatic mutations such as small single-nucleotide variations (SNVs) and small insertions/deletions based on short-read sequencing methods ([63]10), which have intrinsic drawbacks on detecting mutations on longer sequence scope. In addition, a considerable portion of neoepitopes originates from unannotated proteins, a type of brand-new proteins that may result from alternative splicing, nontraditional start sites, or read-through translation events and have not been documented yet in current protein databases ([64]11, [65]12). Unannotated proteins represent a more complex landscape of somatic mutations and can contribute to the diversity of potential neoepitopes for immunotherapy targets. Consequently, the spotlight has now shifted toward innovative methods that can seamlessly capture this broader spectrum of mutations. Full-length ribosome–nascent chain complex–bound mRNA sequencing (FL-RNC seq) allows for the identification of mRNA transcripts that are actively being translated by ribosomes, enabling researchers to achieve a more holistic alignment with the proteome and mitigating the inaccuracies originated from selective translation events ([66]13, [67]14). Moreover, FL-RNC seq can accurately identify the large-scale transcript variants (LSTVs) originating from large insertions, deletions, or rearrangements in both the coding sequence of a gene and noncoding region. Therefore, FL-RNC seq is a valuable tool for breaking the bottleneck of neoepitope discovery. We applied FL-RNC seq on tumor tissues and matched control tissues from the MC38 mouse syngeneic model and identified a pool of LSTV, based on which we further predicted MHC-I– and MHC-II–restricted epitopes using computational predictive tools named FIONA2. We developed an mRNA therapeutic vaccine based on selected LSTV sequences rich in MHC-I– and MHC-II–restricted epitopes. The mRNA vaccine formulated as mRNA lipid nanoparticle (LNP) elicited notable cellular immune response and demonstrated strong tumor inhibition efficacy. When used in synergy with PD-1 blockade, the efficacy was further augmented, suggesting a potential paradigm shift in colorectal cancer therapy. Our findings also suggest that immunogenic MHC class II–restricted neoepitopes may dictate the vaccine efficacy. Further investigation using advanced techniques such as single-cell RNA sequencing (scRNA-seq) unveiled notable rejuvenation in the tumor microenvironment after vaccination. LSTV-based mRNA vaccine generated a durable T cell memory response and prevented tumor relapse efficiently. These findings unveiled that the potential of mRNA vaccines based on LSTV-derived neoepitopes may have substantial implications on the development of personalized RNA vaccine for human cancers. RESULTS LSTVs serve as crucial sources of tumor neoepitopes To determine tumor suppression potential of neoepitopes generated from LSTVs, we selected MC38 tumor model because previous studies have provided ample evidence that MC38 mice have several well-defined neoepitopes originating from point mutations detected by short-read sequencing ([68]9). This allows us to simultaneously use LSTV-derived neoepitopes for comparison in terms of quantity and quality. We harvested tumors from C57BL/6J mice subcutaneously inoculated with MC38 cells. Ribosome–nascent chain complexes (RNCs) were enriched from these tumors and matched normal tissues, enabling comprehensive neoepitope identification through RNC-sequencing (RNC-seq) ([69]15). Data analysis was performed using the GRCm39 genome and transcripts as references.